fs/xfs/libxfs/xfs_bmap.c

   1 /*
   2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_bit.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_defer.h"
  28 #include "xfs_da_format.h"
  29 #include "xfs_da_btree.h"
  30 #include "xfs_dir2.h"
  31 #include "xfs_inode.h"
  32 #include "xfs_btree.h"
  33 #include "xfs_trans.h"
  34 #include "xfs_inode_item.h"
  35 #include "xfs_extfree_item.h"
  36 #include "xfs_alloc.h"
  37 #include "xfs_bmap.h"
  38 #include "xfs_bmap_util.h"
  39 #include "xfs_bmap_btree.h"
  40 #include "xfs_rtalloc.h"
  41 #include "xfs_error.h"
  42 #include "xfs_quota.h"
  43 #include "xfs_trans_space.h"
  44 #include "xfs_buf_item.h"
  45 #include "xfs_trace.h"
  46 #include "xfs_symlink.h"
  47 #include "xfs_attr_leaf.h"
  48 #include "xfs_filestream.h"
  49 #include "xfs_rmap.h"
  50 #include "xfs_ag_resv.h"
  51 #include "xfs_refcount.h"
  52 #include "xfs_rmap_btree.h"
  53 #include "xfs_icache.h"
  54
  55
  56 kmem_zone_t             *xfs_bmap_free_item_zone;
  57
  58 /*
  59  * Miscellaneous helper functions
  60  */
  61
  62 /*
  63  * Compute and fill in the value of the maximum depth of a bmap btree
  64  * in this filesystem.  Done once, during mount.
  65  */
  66 void
  67 xfs_bmap_compute_maxlevels(
  68         xfs_mount_t     *mp,            /* file system mount structure */
  69         int             whichfork)      /* data or attr fork */
  70 {
  71         int             level;          /* btree level */
  72         uint            maxblocks;      /* max blocks at this level */
  73         uint            maxleafents;    /* max leaf entries possible */
  74         int             maxrootrecs;    /* max records in root block */
  75         int             minleafrecs;    /* min records in leaf block */
  76         int             minnoderecs;    /* min records in node block */
  77         int             sz;             /* root block size */
  78
  79         /*
  80          * The maximum number of extents in a file, hence the maximum
  81          * number of leaf entries, is controlled by the type of di_nextents
  82          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
  83          * (a signed 16-bit number, xfs_aextnum_t).
  84          *
  85          * Note that we can no longer assume that if we are in ATTR1 that
  86          * the fork offset of all the inodes will be
  87          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
  88          * with ATTR2 and then mounted back with ATTR1, keeping the
  89          * di_forkoff's fixed but probably at various positions. Therefore,
  90          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
  91          * of a minimum size available.
  92          */
  93         if (whichfork == XFS_DATA_FORK) {
  94                 maxleafents = MAXEXTNUM;
  95                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
  96         } else {
  97                 maxleafents = MAXAEXTNUM;
  98                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
  99         }
 100         maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
 101         minleafrecs = mp->m_bmap_dmnr[0];
 102         minnoderecs = mp->m_bmap_dmnr[1];
 103         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
 104         for (level = 1; maxblocks > 1; level++) {
 105                 if (maxblocks <= maxrootrecs)
 106                         maxblocks = 1;
 107                 else
 108                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
 109         }
 110         mp->m_bm_maxlevels[whichfork] = level;
 111 }
 112
 113 STATIC int                              /* error */
 114 xfs_bmbt_lookup_eq(
 115         struct xfs_btree_cur    *cur,
 116         xfs_fileoff_t           off,
 117         xfs_fsblock_t           bno,
 118         xfs_filblks_t           len,
 119         int                     *stat)  /* success/failure */
 120 {
 121         cur->bc_rec.b.br_startoff = off;
 122         cur->bc_rec.b.br_startblock = bno;
 123         cur->bc_rec.b.br_blockcount = len;
 124         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 125 }
 126
 127 STATIC int                              /* error */
 128 xfs_bmbt_lookup_ge(
 129         struct xfs_btree_cur    *cur,
 130         xfs_fileoff_t           off,
 131         xfs_fsblock_t           bno,
 132         xfs_filblks_t           len,
 133         int                     *stat)  /* success/failure */
 134 {
 135         cur->bc_rec.b.br_startoff = off;
 136         cur->bc_rec.b.br_startblock = bno;
 137         cur->bc_rec.b.br_blockcount = len;
 138         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 139 }
 140
 141 /*
 142  * Check if the inode needs to be converted to btree format.
 143  */
 144 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 145 {
 146         return whichfork != XFS_COW_FORK &&
 147                 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 148                 XFS_IFORK_NEXTENTS(ip, whichfork) >
 149                         XFS_IFORK_MAXEXT(ip, whichfork);
 150 }
 151
 152 /*
 153  * Check if the inode should be converted to extent format.
 154  */
 155 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 156 {
 157         return whichfork != XFS_COW_FORK &&
 158                 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 159                 XFS_IFORK_NEXTENTS(ip, whichfork) <=
 160                         XFS_IFORK_MAXEXT(ip, whichfork);
 161 }
 162
 163 /*
 164  * Update the record referred to by cur to the value given
 165  * by [off, bno, len, state].
 166  * This either works (return 0) or gets an EFSCORRUPTED error.
 167  */
 168 STATIC int
 169 xfs_bmbt_update(
 170         struct xfs_btree_cur    *cur,
 171         xfs_fileoff_t           off,
 172         xfs_fsblock_t           bno,
 173         xfs_filblks_t           len,
 174         xfs_exntst_t            state)
 175 {
 176         union xfs_btree_rec     rec;
 177
 178         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 179         return xfs_btree_update(cur, &rec);
 180 }
 181
 182 /*
 183  * Compute the worst-case number of indirect blocks that will be used
 184  * for ip's delayed extent of length "len".
 185  */
 186 STATIC xfs_filblks_t
 187 xfs_bmap_worst_indlen(
 188         xfs_inode_t     *ip,            /* incore inode pointer */
 189         xfs_filblks_t   len)            /* delayed extent length */
 190 {
 191         int             level;          /* btree level number */
 192         int             maxrecs;        /* maximum record count at this level */
 193         xfs_mount_t     *mp;            /* mount structure */
 194         xfs_filblks_t   rval;           /* return value */
 195         xfs_filblks_t   orig_len;
 196
 197         mp = ip->i_mount;
 198
 199         /* Calculate the worst-case size of the bmbt. */
 200         orig_len = len;
 201         maxrecs = mp->m_bmap_dmxr[0];
 202         for (level = 0, rval = 0;
 203              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
 204              level++) {
 205                 len += maxrecs - 1;
 206                 do_div(len, maxrecs);
 207                 rval += len;
 208                 if (len == 1) {
 209                         rval += XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
 210                                 level - 1;
 211                         break;
 212                 }
 213                 if (level == 0)
 214                         maxrecs = mp->m_bmap_dmxr[1];
 215         }
 216
 217         /* Calculate the worst-case size of the rmapbt. */
 218         if (xfs_sb_version_hasrmapbt(&mp->m_sb))
 219                 rval += 1 + xfs_rmapbt_calc_size(mp, orig_len) +
 220                                 mp->m_rmap_maxlevels;
 221
 222         return rval;
 223 }
 224
 225 /*
 226  * Calculate the default attribute fork offset for newly created inodes.
 227  */
 228 uint
 229 xfs_default_attroffset(
 230         struct xfs_inode        *ip)
 231 {
 232         struct xfs_mount        *mp = ip->i_mount;
 233         uint                    offset;
 234
 235         if (mp->m_sb.sb_inodesize == 256) {
 236                 offset = XFS_LITINO(mp, ip->i_d.di_version) -
 237                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
 238         } else {
 239                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
 240         }
 241
 242         ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
 243         return offset;
 244 }
 245
 246 /*
 247  * Helper routine to reset inode di_forkoff field when switching
 248  * attribute fork from local to extent format - we reset it where
 249  * possible to make space available for inline data fork extents.
 250  */
 251 STATIC void
 252 xfs_bmap_forkoff_reset(
 253         xfs_inode_t     *ip,
 254         int             whichfork)
 255 {
 256         if (whichfork == XFS_ATTR_FORK &&
 257             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
 258             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
 259             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
 260                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
 261
 262                 if (dfl_forkoff > ip->i_d.di_forkoff)
 263                         ip->i_d.di_forkoff = dfl_forkoff;
 264         }
 265 }
 266
 267 #ifdef DEBUG
 268 STATIC struct xfs_buf *
 269 xfs_bmap_get_bp(
 270         struct xfs_btree_cur    *cur,
 271         xfs_fsblock_t           bno)
 272 {
 273         struct xfs_log_item_desc *lidp;
 274         int                     i;
 275
 276         if (!cur)
 277                 return NULL;
 278
 279         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
 280                 if (!cur->bc_bufs[i])
 281                         break;
 282                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
 283                         return cur->bc_bufs[i];
 284         }
 285
 286         /* Chase down all the log items to see if the bp is there */
 287         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
 288                 struct xfs_buf_log_item *bip;
 289                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
 290                 if (bip->bli_item.li_type == XFS_LI_BUF &&
 291                     XFS_BUF_ADDR(bip->bli_buf) == bno)
 292                         return bip->bli_buf;
 293         }
 294
 295         return NULL;
 296 }
 297
 298 STATIC void
 299 xfs_check_block(
 300         struct xfs_btree_block  *block,
 301         xfs_mount_t             *mp,
 302         int                     root,
 303         short                   sz)
 304 {
 305         int                     i, j, dmxr;
 306         __be64                  *pp, *thispa;   /* pointer to block address */
 307         xfs_bmbt_key_t          *prevp, *keyp;
 308
 309         ASSERT(be16_to_cpu(block->bb_level) > 0);
 310
 311         prevp = NULL;
 312         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
 313                 dmxr = mp->m_bmap_dmxr[0];
 314                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
 315
 316                 if (prevp) {
 317                         ASSERT(be64_to_cpu(prevp->br_startoff) <
 318                                be64_to_cpu(keyp->br_startoff));
 319                 }
 320                 prevp = keyp;
 321
 322                 /*
 323                  * Compare the block numbers to see if there are dups.
 324                  */
 325                 if (root)
 326                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
 327                 else
 328                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
 329
 330                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
 331                         if (root)
 332                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
 333                         else
 334                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
 335                         if (*thispa == *pp) {
 336                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
 337                                         __func__, j, i,
 338                                         (unsigned long long)be64_to_cpu(*thispa));
 339                                 panic("%s: ptrs are equal in node\n",
 340                                         __func__);
 341                         }
 342                 }
 343         }
 344 }
 345
 346 /*
 347  * Check that the extents for the inode ip are in the right order in all
 348  * btree leaves. THis becomes prohibitively expensive for large extent count
 349  * files, so don't bother with inodes that have more than 10,000 extents in
 350  * them. The btree record ordering checks will still be done, so for such large
 351  * bmapbt constructs that is going to catch most corruptions.
 352  */
 353 STATIC void
 354 xfs_bmap_check_leaf_extents(
 355         xfs_btree_cur_t         *cur,   /* btree cursor or null */
 356         xfs_inode_t             *ip,            /* incore inode pointer */
 357         int                     whichfork)      /* data or attr fork */
 358 {
 359         struct xfs_btree_block  *block; /* current btree block */
 360         xfs_fsblock_t           bno;    /* block # of "block" */
 361         xfs_buf_t               *bp;    /* buffer for "block" */
 362         int                     error;  /* error return value */
 363         xfs_extnum_t            i=0, j; /* index into the extents list */
 364         xfs_ifork_t             *ifp;   /* fork structure */
 365         int                     level;  /* btree level, for checking */
 366         xfs_mount_t             *mp;    /* file system mount structure */
 367         __be64                  *pp;    /* pointer to block address */
 368         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
 369         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
 370         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
 371         int                     bp_release = 0;
 372
 373         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
 374                 return;
 375         }
 376
 377         /* skip large extent count inodes */
 378         if (ip->i_d.di_nextents > 10000)
 379                 return;
 380
 381         bno = NULLFSBLOCK;
 382         mp = ip->i_mount;
 383         ifp = XFS_IFORK_PTR(ip, whichfork);
 384         block = ifp->if_broot;
 385         /*
 386          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 387          */
 388         level = be16_to_cpu(block->bb_level);
 389         ASSERT(level > 0);
 390         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
 391         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 392         bno = be64_to_cpu(*pp);
 393
 394         ASSERT(bno != NULLFSBLOCK);
 395         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 396         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 397
 398         /*
 399          * Go down the tree until leaf level is reached, following the first
 400          * pointer (leftmost) at each level.
 401          */
 402         while (level-- > 0) {
 403                 /* See if buf is in cur first */
 404                 bp_release = 0;
 405                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 406                 if (!bp) {
 407                         bp_release = 1;
 408                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 409                                                 XFS_BMAP_BTREE_REF,
 410                                                 &xfs_bmbt_buf_ops);
 411                         if (error)
 412                                 goto error_norelse;
 413                 }
 414                 block = XFS_BUF_TO_BLOCK(bp);
 415                 if (level == 0)
 416                         break;
 417
 418                 /*
 419                  * Check this block for basic sanity (increasing keys and
 420                  * no duplicate blocks).
 421                  */
 422
 423                 xfs_check_block(block, mp, 0, 0);
 424                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 425                 bno = be64_to_cpu(*pp);
 426                 XFS_WANT_CORRUPTED_GOTO(mp,
 427                                         XFS_FSB_SANITY_CHECK(mp, bno), error0);
 428                 if (bp_release) {
 429                         bp_release = 0;
 430                         xfs_trans_brelse(NULL, bp);
 431                 }
 432         }
 433
 434         /*
 435          * Here with bp and block set to the leftmost leaf node in the tree.
 436          */
 437         i = 0;
 438
 439         /*
 440          * Loop over all leaf nodes checking that all extents are in the right order.
 441          */
 442         for (;;) {
 443                 xfs_fsblock_t   nextbno;
 444                 xfs_extnum_t    num_recs;
 445
 446
 447                 num_recs = xfs_btree_get_numrecs(block);
 448
 449                 /*
 450                  * Read-ahead the next leaf block, if any.
 451                  */
 452
 453                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 454
 455                 /*
 456                  * Check all the extents to make sure they are OK.
 457                  * If we had a previous block, the last entry should
 458                  * conform with the first entry in this one.
 459                  */
 460
 461                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
 462                 if (i) {
 463                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
 464                                xfs_bmbt_disk_get_blockcount(&last) <=
 465                                xfs_bmbt_disk_get_startoff(ep));
 466                 }
 467                 for (j = 1; j < num_recs; j++) {
 468                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
 469                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
 470                                xfs_bmbt_disk_get_blockcount(ep) <=
 471                                xfs_bmbt_disk_get_startoff(nextp));
 472                         ep = nextp;
 473                 }
 474
 475                 last = *ep;
 476                 i += num_recs;
 477                 if (bp_release) {
 478                         bp_release = 0;
 479                         xfs_trans_brelse(NULL, bp);
 480                 }
 481                 bno = nextbno;
 482                 /*
 483                  * If we've reached the end, stop.
 484                  */
 485                 if (bno == NULLFSBLOCK)
 486                         break;
 487
 488                 bp_release = 0;
 489                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 490                 if (!bp) {
 491                         bp_release = 1;
 492                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 493                                                 XFS_BMAP_BTREE_REF,
 494                                                 &xfs_bmbt_buf_ops);
 495                         if (error)
 496                                 goto error_norelse;
 497                 }
 498                 block = XFS_BUF_TO_BLOCK(bp);
 499         }
 500
 501         return;
 502
 503 error0:
 504         xfs_warn(mp, "%s: at error0", __func__);
 505         if (bp_release)
 506                 xfs_trans_brelse(NULL, bp);
 507 error_norelse:
 508         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
 509                 __func__, i);
 510         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
 511         return;
 512 }
 513
 514 /*
 515  * Add bmap trace insert entries for all the contents of the extent records.
 516  */
 517 void
 518 xfs_bmap_trace_exlist(
 519         xfs_inode_t     *ip,            /* incore inode pointer */
 520         xfs_extnum_t    cnt,            /* count of entries in the list */
 521         int             whichfork,      /* data or attr or cow fork */
 522         unsigned long   caller_ip)
 523 {
 524         xfs_extnum_t    idx;            /* extent record index */
 525         xfs_ifork_t     *ifp;           /* inode fork pointer */
 526         int             state = 0;
 527
 528         if (whichfork == XFS_ATTR_FORK)
 529                 state |= BMAP_ATTRFORK;
 530         else if (whichfork == XFS_COW_FORK)
 531                 state |= BMAP_COWFORK;
 532
 533         ifp = XFS_IFORK_PTR(ip, whichfork);
 534         ASSERT(cnt == xfs_iext_count(ifp));
 535         for (idx = 0; idx < cnt; idx++)
 536                 trace_xfs_extlist(ip, idx, state, caller_ip);
 537 }
 538
 539 /*
 540  * Validate that the bmbt_irecs being returned from bmapi are valid
 541  * given the caller's original parameters.  Specifically check the
 542  * ranges of the returned irecs to ensure that they only extend beyond
 543  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
 544  */
 545 STATIC void
 546 xfs_bmap_validate_ret(
 547         xfs_fileoff_t           bno,
 548         xfs_filblks_t           len,
 549         int                     flags,
 550         xfs_bmbt_irec_t         *mval,
 551         int                     nmap,
 552         int                     ret_nmap)
 553 {
 554         int                     i;              /* index to map values */
 555
 556         ASSERT(ret_nmap <= nmap);
 557
 558         for (i = 0; i < ret_nmap; i++) {
 559                 ASSERT(mval[i].br_blockcount > 0);
 560                 if (!(flags & XFS_BMAPI_ENTIRE)) {
 561                         ASSERT(mval[i].br_startoff >= bno);
 562                         ASSERT(mval[i].br_blockcount <= len);
 563                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
 564                                bno + len);
 565                 } else {
 566                         ASSERT(mval[i].br_startoff < bno + len);
 567                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
 568                                bno);
 569                 }
 570                 ASSERT(i == 0 ||
 571                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
 572                        mval[i].br_startoff);
 573                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
 574                        mval[i].br_startblock != HOLESTARTBLOCK);
 575                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
 576                        mval[i].br_state == XFS_EXT_UNWRITTEN);
 577         }
 578 }
 579
 580 #else
 581 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
 582 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 583 #endif /* DEBUG */
 584
 585 /*
 586  * bmap free list manipulation functions
 587  */
 588
 589 /*
 590  * Add the extent to the list of extents to be free at transaction end.
 591  * The list is maintained sorted (by block number).
 592  */
 593 void
 594 xfs_bmap_add_free(
 595         struct xfs_mount                *mp,
 596         struct xfs_defer_ops            *dfops,
 597         xfs_fsblock_t                   bno,
 598         xfs_filblks_t                   len,
 599         struct xfs_owner_info           *oinfo)
 600 {
 601         struct xfs_extent_free_item     *new;           /* new element */
 602 #ifdef DEBUG
 603         xfs_agnumber_t          agno;
 604         xfs_agblock_t           agbno;
 605
 606         ASSERT(bno != NULLFSBLOCK);
 607         ASSERT(len > 0);
 608         ASSERT(len <= MAXEXTLEN);
 609         ASSERT(!isnullstartblock(bno));
 610         agno = XFS_FSB_TO_AGNO(mp, bno);
 611         agbno = XFS_FSB_TO_AGBNO(mp, bno);
 612         ASSERT(agno < mp->m_sb.sb_agcount);
 613         ASSERT(agbno < mp->m_sb.sb_agblocks);
 614         ASSERT(len < mp->m_sb.sb_agblocks);
 615         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
 616 #endif
 617         ASSERT(xfs_bmap_free_item_zone != NULL);
 618
 619         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
 620         new->xefi_startblock = bno;
 621         new->xefi_blockcount = (xfs_extlen_t)len;
 622         if (oinfo)
 623                 new->xefi_oinfo = *oinfo;
 624         else
 625                 xfs_rmap_skip_owner_update(&new->xefi_oinfo);
 626         trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0,
 627                         XFS_FSB_TO_AGBNO(mp, bno), len);
 628         xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
 629 }
 630
 631 /*
 632  * Inode fork format manipulation functions
 633  */
 634
 635 /*
 636  * Transform a btree format file with only one leaf node, where the
 637  * extents list will fit in the inode, into an extents format file.
 638  * Since the file extents are already in-core, all we have to do is
 639  * give up the space for the btree root and pitch the leaf block.
 640  */
 641 STATIC int                              /* error */
 642 xfs_bmap_btree_to_extents(
 643         xfs_trans_t             *tp,    /* transaction pointer */
 644         xfs_inode_t             *ip,    /* incore inode pointer */
 645         xfs_btree_cur_t         *cur,   /* btree cursor */
 646         int                     *logflagsp, /* inode logging flags */
 647         int                     whichfork)  /* data or attr fork */
 648 {
 649         /* REFERENCED */
 650         struct xfs_btree_block  *cblock;/* child btree block */
 651         xfs_fsblock_t           cbno;   /* child block number */
 652         xfs_buf_t               *cbp;   /* child block's buffer */
 653         int                     error;  /* error return value */
 654         xfs_ifork_t             *ifp;   /* inode fork data */
 655         xfs_mount_t             *mp;    /* mount point structure */
 656         __be64                  *pp;    /* ptr to block address */
 657         struct xfs_btree_block  *rblock;/* root btree block */
 658         struct xfs_owner_info   oinfo;
 659
 660         mp = ip->i_mount;
 661         ifp = XFS_IFORK_PTR(ip, whichfork);
 662         ASSERT(whichfork != XFS_COW_FORK);
 663         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
 664         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
 665         rblock = ifp->if_broot;
 666         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
 667         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
 668         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
 669         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
 670         cbno = be64_to_cpu(*pp);
 671         *logflagsp = 0;
 672 #ifdef DEBUG
 673         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
 674                 return error;
 675 #endif
 676         error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
 677                                 &xfs_bmbt_buf_ops);
 678         if (error)
 679                 return error;
 680         cblock = XFS_BUF_TO_BLOCK(cbp);
 681         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
 682                 return error;
 683         xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
 684         xfs_bmap_add_free(mp, cur->bc_private.b.dfops, cbno, 1, &oinfo);
 685         ip->i_d.di_nblocks--;
 686         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 687         xfs_trans_binval(tp, cbp);
 688         if (cur->bc_bufs[0] == cbp)
 689                 cur->bc_bufs[0] = NULL;
 690         xfs_iroot_realloc(ip, -1, whichfork);
 691         ASSERT(ifp->if_broot == NULL);
 692         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
 693         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 694         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 695         return 0;
 696 }
 697
 698 /*
 699  * Convert an extents-format file into a btree-format file.
 700  * The new file will have a root block (in the inode) and a single child block.
 701  */
 702 STATIC int                                      /* error */
 703 xfs_bmap_extents_to_btree(
 704         xfs_trans_t             *tp,            /* transaction pointer */
 705         xfs_inode_t             *ip,            /* incore inode pointer */
 706         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 707         struct xfs_defer_ops    *dfops,         /* blocks freed in xaction */
 708         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 709         int                     wasdel,         /* converting a delayed alloc */
 710         int                     *logflagsp,     /* inode logging flags */
 711         int                     whichfork)      /* data or attr fork */
 712 {
 713         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
 714         xfs_buf_t               *abp;           /* buffer for ablock */
 715         xfs_alloc_arg_t         args;           /* allocation arguments */
 716         xfs_bmbt_rec_t          *arp;           /* child record pointer */
 717         struct xfs_btree_block  *block;         /* btree root block */
 718         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 719         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
 720         int                     error;          /* error return value */
 721         xfs_extnum_t            i, cnt;         /* extent record index */
 722         xfs_ifork_t             *ifp;           /* inode fork pointer */
 723         xfs_bmbt_key_t          *kp;            /* root block key pointer */
 724         xfs_mount_t             *mp;            /* mount structure */
 725         xfs_extnum_t            nextents;       /* number of file extents */
 726         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
 727
 728         mp = ip->i_mount;
 729         ASSERT(whichfork != XFS_COW_FORK);
 730         ifp = XFS_IFORK_PTR(ip, whichfork);
 731         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
 732
 733         /*
 734          * Make space in the inode incore.
 735          */
 736         xfs_iroot_realloc(ip, 1, whichfork);
 737         ifp->if_flags |= XFS_IFBROOT;
 738
 739         /*
 740          * Fill in the root.
 741          */
 742         block = ifp->if_broot;
 743         xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 744                                  XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
 745                                  XFS_BTREE_LONG_PTRS);
 746         /*
 747          * Need a cursor.  Can't allocate until bb_level is filled in.
 748          */
 749         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
 750         cur->bc_private.b.firstblock = *firstblock;
 751         cur->bc_private.b.dfops = dfops;
 752         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
 753         /*
 754          * Convert to a btree with two levels, one record in root.
 755          */
 756         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
 757         memset(&args, 0, sizeof(args));
 758         args.tp = tp;
 759         args.mp = mp;
 760         xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
 761         args.firstblock = *firstblock;
 762         if (*firstblock == NULLFSBLOCK) {
 763                 args.type = XFS_ALLOCTYPE_START_BNO;
 764                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
 765         } else if (dfops->dop_low) {
 766                 args.type = XFS_ALLOCTYPE_START_BNO;
 767                 args.fsbno = *firstblock;
 768         } else {
 769                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 770                 args.fsbno = *firstblock;
 771         }
 772         args.minlen = args.maxlen = args.prod = 1;
 773         args.wasdel = wasdel;
 774         *logflagsp = 0;
 775         if ((error = xfs_alloc_vextent(&args))) {
 776                 xfs_iroot_realloc(ip, -1, whichfork);
 777                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 778                 return error;
 779         }
 780
 781         if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
 782                 xfs_iroot_realloc(ip, -1, whichfork);
 783                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 784                 return -ENOSPC;
 785         }
 786         /*
 787          * Allocation can't fail, the space was reserved.
 788          */
 789         ASSERT(*firstblock == NULLFSBLOCK ||
 790                args.agno >= XFS_FSB_TO_AGNO(mp, *firstblock));
 791         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
 792         cur->bc_private.b.allocated++;
 793         ip->i_d.di_nblocks++;
 794         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
 795         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
 796         /*
 797          * Fill in the child block.
 798          */
 799         abp->b_ops = &xfs_bmbt_buf_ops;
 800         ablock = XFS_BUF_TO_BLOCK(abp);
 801         xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 802                                 XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
 803                                 XFS_BTREE_LONG_PTRS);
 804
 805         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 806         nextents =  xfs_iext_count(ifp);
 807         for (cnt = i = 0; i < nextents; i++) {
 808                 ep = xfs_iext_get_ext(ifp, i);
 809                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
 810                         arp->l0 = cpu_to_be64(ep->l0);
 811                         arp->l1 = cpu_to_be64(ep->l1);
 812                         arp++; cnt++;
 813                 }
 814         }
 815         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
 816         xfs_btree_set_numrecs(ablock, cnt);
 817
 818         /*
 819          * Fill in the root key and pointer.
 820          */
 821         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
 822         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 823         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
 824         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
 825                                                 be16_to_cpu(block->bb_level)));
 826         *pp = cpu_to_be64(args.fsbno);
 827
 828         /*
 829          * Do all this logging at the end so that
 830          * the root is at the right level.
 831          */
 832         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
 833         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
 834         ASSERT(*curp == NULL);
 835         *curp = cur;
 836         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
 837         return 0;
 838 }
 839
 840 /*
 841  * Convert a local file to an extents file.
 842  * This code is out of bounds for data forks of regular files,
 843  * since the file data needs to get logged so things will stay consistent.
 844  * (The bmap-level manipulations are ok, though).
 845  */
 846 void
 847 xfs_bmap_local_to_extents_empty(
 848         struct xfs_inode        *ip,
 849         int                     whichfork)
 850 {
 851         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 852
 853         ASSERT(whichfork != XFS_COW_FORK);
 854         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 855         ASSERT(ifp->if_bytes == 0);
 856         ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
 857
 858         xfs_bmap_forkoff_reset(ip, whichfork);
 859         ifp->if_flags &= ~XFS_IFINLINE;
 860         ifp->if_flags |= XFS_IFEXTENTS;
 861         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 862 }
 863
 864
 865 STATIC int                              /* error */
 866 xfs_bmap_local_to_extents(
 867         xfs_trans_t     *tp,            /* transaction pointer */
 868         xfs_inode_t     *ip,            /* incore inode pointer */
 869         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 870         xfs_extlen_t    total,          /* total blocks needed by transaction */
 871         int             *logflagsp,     /* inode logging flags */
 872         int             whichfork,
 873         void            (*init_fn)(struct xfs_trans *tp,
 874                                    struct xfs_buf *bp,
 875                                    struct xfs_inode *ip,
 876                                    struct xfs_ifork *ifp))
 877 {
 878         int             error = 0;
 879         int             flags;          /* logging flags returned */
 880         xfs_ifork_t     *ifp;           /* inode fork pointer */
 881         xfs_alloc_arg_t args;           /* allocation arguments */
 882         xfs_buf_t       *bp;            /* buffer for extent block */
 883         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
 884
 885         /*
 886          * We don't want to deal with the case of keeping inode data inline yet.
 887          * So sending the data fork of a regular inode is invalid.
 888          */
 889         ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
 890         ifp = XFS_IFORK_PTR(ip, whichfork);
 891         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 892
 893         if (!ifp->if_bytes) {
 894                 xfs_bmap_local_to_extents_empty(ip, whichfork);
 895                 flags = XFS_ILOG_CORE;
 896                 goto done;
 897         }
 898
 899         flags = 0;
 900         error = 0;
 901         ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
 902                                                                 XFS_IFINLINE);
 903         memset(&args, 0, sizeof(args));
 904         args.tp = tp;
 905         args.mp = ip->i_mount;
 906         xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
 907         args.firstblock = *firstblock;
 908         /*
 909          * Allocate a block.  We know we need only one, since the
 910          * file currently fits in an inode.
 911          */
 912         if (*firstblock == NULLFSBLOCK) {
 913                 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
 914                 args.type = XFS_ALLOCTYPE_START_BNO;
 915         } else {
 916                 args.fsbno = *firstblock;
 917                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 918         }
 919         args.total = total;
 920         args.minlen = args.maxlen = args.prod = 1;
 921         error = xfs_alloc_vextent(&args);
 922         if (error)
 923                 goto done;
 924
 925         /* Can't fail, the space was reserved. */
 926         ASSERT(args.fsbno != NULLFSBLOCK);
 927         ASSERT(args.len == 1);
 928         *firstblock = args.fsbno;
 929         bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
 930
 931         /*
 932          * Initialize the block, copy the data and log the remote buffer.
 933          *
 934          * The callout is responsible for logging because the remote format
 935          * might differ from the local format and thus we don't know how much to
 936          * log here. Note that init_fn must also set the buffer log item type
 937          * correctly.
 938          */
 939         init_fn(tp, bp, ip, ifp);
 940
 941         /* account for the change in fork size */
 942         xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
 943         xfs_bmap_local_to_extents_empty(ip, whichfork);
 944         flags |= XFS_ILOG_CORE;
 945
 946         xfs_iext_add(ifp, 0, 1);
 947         ep = xfs_iext_get_ext(ifp, 0);
 948         xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
 949         trace_xfs_bmap_post_update(ip, 0,
 950                         whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
 951                         _THIS_IP_);
 952         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 953         ip->i_d.di_nblocks = 1;
 954         xfs_trans_mod_dquot_byino(tp, ip,
 955                 XFS_TRANS_DQ_BCOUNT, 1L);
 956         flags |= xfs_ilog_fext(whichfork);
 957
 958 done:
 959         *logflagsp = flags;
 960         return error;
 961 }
 962
 963 /*
 964  * Called from xfs_bmap_add_attrfork to handle btree format files.
 965  */
 966 STATIC int                                      /* error */
 967 xfs_bmap_add_attrfork_btree(
 968         xfs_trans_t             *tp,            /* transaction pointer */
 969         xfs_inode_t             *ip,            /* incore inode pointer */
 970         xfs_fsblock_t           *firstblock,    /* first block allocated */
 971         struct xfs_defer_ops    *dfops,         /* blocks to free at commit */
 972         int                     *flags)         /* inode logging flags */
 973 {
 974         xfs_btree_cur_t         *cur;           /* btree cursor */
 975         int                     error;          /* error return value */
 976         xfs_mount_t             *mp;            /* file system mount struct */
 977         int                     stat;           /* newroot status */
 978
 979         mp = ip->i_mount;
 980         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 981                 *flags |= XFS_ILOG_DBROOT;
 982         else {
 983                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 984                 cur->bc_private.b.dfops = dfops;
 985                 cur->bc_private.b.firstblock = *firstblock;
 986                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 987                         goto error0;
 988                 /* must be at least one entry */
 989                 XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
 990                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 991                         goto error0;
 992                 if (stat == 0) {
 993                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 994                         return -ENOSPC;
 995                 }
 996                 *firstblock = cur->bc_private.b.firstblock;
 997                 cur->bc_private.b.allocated = 0;
 998                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 999         }
1000         return 0;
1001 error0:
1002         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1003         return error;
1004 }
1005
1006 /*
1007  * Called from xfs_bmap_add_attrfork to handle extents format files.
1008  */
1009 STATIC int                                      /* error */
1010 xfs_bmap_add_attrfork_extents(
1011         xfs_trans_t             *tp,            /* transaction pointer */
1012         xfs_inode_t             *ip,            /* incore inode pointer */
1013         xfs_fsblock_t           *firstblock,    /* first block allocated */
1014         struct xfs_defer_ops    *dfops,         /* blocks to free at commit */
1015         int                     *flags)         /* inode logging flags */
1016 {
1017         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
1018         int                     error;          /* error return value */
1019
1020         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1021                 return 0;
1022         cur = NULL;
1023         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, &cur, 0,
1024                 flags, XFS_DATA_FORK);
1025         if (cur) {
1026                 cur->bc_private.b.allocated = 0;
1027                 xfs_btree_del_cursor(cur,
1028                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1029         }
1030         return error;
1031 }
1032
1033 /*
1034  * Called from xfs_bmap_add_attrfork to handle local format files. Each
1035  * different data fork content type needs a different callout to do the
1036  * conversion. Some are basic and only require special block initialisation
1037  * callouts for the data formating, others (directories) are so specialised they
1038  * handle everything themselves.
1039  *
1040  * XXX (dgc): investigate whether directory conversion can use the generic
1041  * formatting callout. It should be possible - it's just a very complex
1042  * formatter.
1043  */
1044 STATIC int                                      /* error */
1045 xfs_bmap_add_attrfork_local(
1046         xfs_trans_t             *tp,            /* transaction pointer */
1047         xfs_inode_t             *ip,            /* incore inode pointer */
1048         xfs_fsblock_t           *firstblock,    /* first block allocated */
1049         struct xfs_defer_ops    *dfops,         /* blocks to free at commit */
1050         int                     *flags)         /* inode logging flags */
1051 {
1052         xfs_da_args_t           dargs;          /* args for dir/attr code */
1053
1054         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1055                 return 0;
1056
1057         if (S_ISDIR(VFS_I(ip)->i_mode)) {
1058                 memset(&dargs, 0, sizeof(dargs));
1059                 dargs.geo = ip->i_mount->m_dir_geo;
1060                 dargs.dp = ip;
1061                 dargs.firstblock = firstblock;
1062                 dargs.dfops = dfops;
1063                 dargs.total = dargs.geo->fsbcount;
1064                 dargs.whichfork = XFS_DATA_FORK;
1065                 dargs.trans = tp;
1066                 return xfs_dir2_sf_to_block(&dargs);
1067         }
1068
1069         if (S_ISLNK(VFS_I(ip)->i_mode))
1070                 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1071                                                  flags, XFS_DATA_FORK,
1072                                                  xfs_symlink_local_to_remote);
1073
1074         /* should only be called for types that support local format data */
1075         ASSERT(0);
1076         return -EFSCORRUPTED;
1077 }
1078
1079 /*
1080  * Convert inode from non-attributed to attributed.
1081  * Must not be in a transaction, ip must not be locked.
1082  */
1083 int                                             /* error code */
1084 xfs_bmap_add_attrfork(
1085         xfs_inode_t             *ip,            /* incore inode pointer */
1086         int                     size,           /* space new attribute needs */
1087         int                     rsvd)           /* xact may use reserved blks */
1088 {
1089         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
1090         struct xfs_defer_ops    dfops;          /* freed extent records */
1091         xfs_mount_t             *mp;            /* mount structure */
1092         xfs_trans_t             *tp;            /* transaction pointer */
1093         int                     blks;           /* space reservation */
1094         int                     version = 1;    /* superblock attr version */
1095         int                     logflags;       /* logging flags */
1096         int                     error;          /* error return value */
1097
1098         ASSERT(XFS_IFORK_Q(ip) == 0);
1099
1100         mp = ip->i_mount;
1101         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1102
1103         blks = XFS_ADDAFORK_SPACE_RES(mp);
1104
1105         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1106                         rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1107         if (error)
1108                 return error;
1109
1110         xfs_ilock(ip, XFS_ILOCK_EXCL);
1111         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1112                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1113                         XFS_QMOPT_RES_REGBLKS);
1114         if (error)
1115                 goto trans_cancel;
1116         if (XFS_IFORK_Q(ip))
1117                 goto trans_cancel;
1118         if (ip->i_d.di_anextents != 0) {
1119                 error = -EFSCORRUPTED;
1120                 goto trans_cancel;
1121         }
1122         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1123                 /*
1124                  * For inodes coming from pre-6.2 filesystems.
1125                  */
1126                 ASSERT(ip->i_d.di_aformat == 0);
1127                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1128         }
1129
1130         xfs_trans_ijoin(tp, ip, 0);
1131         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1132
1133         switch (ip->i_d.di_format) {
1134         case XFS_DINODE_FMT_DEV:
1135                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1136                 break;
1137         case XFS_DINODE_FMT_UUID:
1138                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1139                 break;
1140         case XFS_DINODE_FMT_LOCAL:
1141         case XFS_DINODE_FMT_EXTENTS:
1142         case XFS_DINODE_FMT_BTREE:
1143                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1144                 if (!ip->i_d.di_forkoff)
1145                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1146                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1147                         version = 2;
1148                 break;
1149         default:
1150                 ASSERT(0);
1151                 error = -EINVAL;
1152                 goto trans_cancel;
1153         }
1154
1155         ASSERT(ip->i_afp == NULL);
1156         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1157         ip->i_afp->if_flags = XFS_IFEXTENTS;
1158         logflags = 0;
1159         xfs_defer_init(&dfops, &firstblock);
1160         switch (ip->i_d.di_format) {
1161         case XFS_DINODE_FMT_LOCAL:
1162                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &dfops,
1163                         &logflags);
1164                 break;
1165         case XFS_DINODE_FMT_EXTENTS:
1166                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1167                         &dfops, &logflags);
1168                 break;
1169         case XFS_DINODE_FMT_BTREE:
1170                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &dfops,
1171                         &logflags);
1172                 break;
1173         default:
1174                 error = 0;
1175                 break;
1176         }
1177         if (logflags)
1178                 xfs_trans_log_inode(tp, ip, logflags);
1179         if (error)
1180                 goto bmap_cancel;
1181         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1182            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1183                 bool log_sb = false;
1184
1185                 spin_lock(&mp->m_sb_lock);
1186                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1187                         xfs_sb_version_addattr(&mp->m_sb);
1188                         log_sb = true;
1189                 }
1190                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1191                         xfs_sb_version_addattr2(&mp->m_sb);
1192                         log_sb = true;
1193                 }
1194                 spin_unlock(&mp->m_sb_lock);
1195                 if (log_sb)
1196                         xfs_log_sb(tp);
1197         }
1198
1199         error = xfs_defer_finish(&tp, &dfops, NULL);
1200         if (error)
1201                 goto bmap_cancel;
1202         error = xfs_trans_commit(tp);
1203         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1204         return error;
1205
1206 bmap_cancel:
1207         xfs_defer_cancel(&dfops);
1208 trans_cancel:
1209         xfs_trans_cancel(tp);
1210         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1211         return error;
1212 }
1213
1214 /*
1215  * Internal and external extent tree search functions.
1216  */
1217
1218 /*
1219  * Read in the extents to if_extents.
1220  * All inode fields are set up by caller, we just traverse the btree
1221  * and copy the records in. If the file system cannot contain unwritten
1222  * extents, the records are checked for no "state" flags.
1223  */
1224 int                                     /* error */
1225 xfs_bmap_read_extents(
1226         xfs_trans_t             *tp,    /* transaction pointer */
1227         xfs_inode_t             *ip,    /* incore inode */
1228         int                     whichfork) /* data or attr fork */
1229 {
1230         struct xfs_btree_block  *block; /* current btree block */
1231         xfs_fsblock_t           bno;    /* block # of "block" */
1232         xfs_buf_t               *bp;    /* buffer for "block" */
1233         int                     error;  /* error return value */
1234         xfs_extnum_t            i, j;   /* index into the extents list */
1235         xfs_ifork_t             *ifp;   /* fork structure */
1236         int                     level;  /* btree level, for checking */
1237         xfs_mount_t             *mp;    /* file system mount structure */
1238         __be64                  *pp;    /* pointer to block address */
1239         /* REFERENCED */
1240         xfs_extnum_t            room;   /* number of entries there's room for */
1241
1242         mp = ip->i_mount;
1243         ifp = XFS_IFORK_PTR(ip, whichfork);
1244         block = ifp->if_broot;
1245         /*
1246          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1247          */
1248         level = be16_to_cpu(block->bb_level);
1249         ASSERT(level > 0);
1250         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1251         bno = be64_to_cpu(*pp);
1252
1253         /*
1254          * Go down the tree until leaf level is reached, following the first
1255          * pointer (leftmost) at each level.
1256          */
1257         while (level-- > 0) {
1258                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1259                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1260                 if (error)
1261                         return error;
1262                 block = XFS_BUF_TO_BLOCK(bp);
1263                 if (level == 0)
1264                         break;
1265                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1266                 bno = be64_to_cpu(*pp);
1267                 XFS_WANT_CORRUPTED_GOTO(mp,
1268                         XFS_FSB_SANITY_CHECK(mp, bno), error0);
1269                 xfs_trans_brelse(tp, bp);
1270         }
1271         /*
1272          * Here with bp and block set to the leftmost leaf node in the tree.
1273          */
1274         room = xfs_iext_count(ifp);
1275         i = 0;
1276         /*
1277          * Loop over all leaf nodes.  Copy information to the extent records.
1278          */
1279         for (;;) {
1280                 xfs_bmbt_rec_t  *frp;
1281                 xfs_fsblock_t   nextbno;
1282                 xfs_extnum_t    num_recs;
1283                 xfs_extnum_t    start;
1284
1285                 num_recs = xfs_btree_get_numrecs(block);
1286                 if (unlikely(i + num_recs > room)) {
1287                         ASSERT(i + num_recs <= room);
1288                         xfs_warn(ip->i_mount,
1289                                 "corrupt dinode %Lu, (btree extents).",
1290                                 (unsigned long long) ip->i_ino);
1291                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1292                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1293                         goto error0;
1294                 }
1295                 /*
1296                  * Read-ahead the next leaf block, if any.
1297                  */
1298                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1299                 if (nextbno != NULLFSBLOCK)
1300                         xfs_btree_reada_bufl(mp, nextbno, 1,
1301                                              &xfs_bmbt_buf_ops);
1302                 /*
1303                  * Copy records into the extent records.
1304                  */
1305                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1306                 start = i;
1307                 for (j = 0; j < num_recs; j++, i++, frp++) {
1308                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1309                         trp->l0 = be64_to_cpu(frp->l0);
1310                         trp->l1 = be64_to_cpu(frp->l1);
1311                         if (!xfs_bmbt_validate_extent(mp, whichfork, trp)) {
1312                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1313                                                  XFS_ERRLEVEL_LOW, mp);
1314                                 goto error0;
1315                         }
1316                 }
1317                 xfs_trans_brelse(tp, bp);
1318                 bno = nextbno;
1319                 /*
1320                  * If we've reached the end, stop.
1321                  */
1322                 if (bno == NULLFSBLOCK)
1323                         break;
1324                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1325                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1326                 if (error)
1327                         return error;
1328                 block = XFS_BUF_TO_BLOCK(bp);
1329         }
1330         if (i != XFS_IFORK_NEXTENTS(ip, whichfork))
1331                 return -EFSCORRUPTED;
1332         ASSERT(i == xfs_iext_count(ifp));
1333         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1334         return 0;
1335 error0:
1336         xfs_trans_brelse(tp, bp);
1337         return -EFSCORRUPTED;
1338 }
1339
1340 /*
1341  * Returns the file-relative block number of the first unused block(s)
1342  * in the file with at least "len" logically contiguous blocks free.
1343  * This is the lowest-address hole if the file has holes, else the first block
1344  * past the end of file.
1345  * Return 0 if the file is currently local (in-inode).
1346  */
1347 int                                             /* error */
1348 xfs_bmap_first_unused(
1349         xfs_trans_t     *tp,                    /* transaction pointer */
1350         xfs_inode_t     *ip,                    /* incore inode */
1351         xfs_extlen_t    len,                    /* size of hole to find */
1352         xfs_fileoff_t   *first_unused,          /* unused block */
1353         int             whichfork)              /* data or attr fork */
1354 {
1355         int             error;                  /* error return value */
1356         int             idx;                    /* extent record index */
1357         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1358         xfs_fileoff_t   lastaddr;               /* last block number seen */
1359         xfs_fileoff_t   lowest;                 /* lowest useful block */
1360         xfs_fileoff_t   max;                    /* starting useful block */
1361         xfs_fileoff_t   off;                    /* offset for this block */
1362         xfs_extnum_t    nextents;               /* number of extent entries */
1363
1364         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1365                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1366                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1367         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1368                 *first_unused = 0;
1369                 return 0;
1370         }
1371         ifp = XFS_IFORK_PTR(ip, whichfork);
1372         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1373             (error = xfs_iread_extents(tp, ip, whichfork)))
1374                 return error;
1375         lowest = *first_unused;
1376         nextents = xfs_iext_count(ifp);
1377         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1378                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1379                 off = xfs_bmbt_get_startoff(ep);
1380                 /*
1381                  * See if the hole before this extent will work.
1382                  */
1383                 if (off >= lowest + len && off - max >= len) {
1384                         *first_unused = max;
1385                         return 0;
1386                 }
1387                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1388                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1389         }
1390         *first_unused = max;
1391         return 0;
1392 }
1393
1394 /*
1395  * Returns the file-relative block number of the last block - 1 before
1396  * last_block (input value) in the file.
1397  * This is not based on i_size, it is based on the extent records.
1398  * Returns 0 for local files, as they do not have extent records.
1399  */
1400 int                                             /* error */
1401 xfs_bmap_last_before(
1402         struct xfs_trans        *tp,            /* transaction pointer */
1403         struct xfs_inode        *ip,            /* incore inode */
1404         xfs_fileoff_t           *last_block,    /* last block */
1405         int                     whichfork)      /* data or attr fork */
1406 {
1407         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1408         struct xfs_bmbt_irec    got;
1409         xfs_extnum_t            idx;
1410         int                     error;
1411
1412         switch (XFS_IFORK_FORMAT(ip, whichfork)) {
1413         case XFS_DINODE_FMT_LOCAL:
1414                 *last_block = 0;
1415                 return 0;
1416         case XFS_DINODE_FMT_BTREE:
1417         case XFS_DINODE_FMT_EXTENTS:
1418                 break;
1419         default:
1420                 return -EIO;
1421         }
1422
1423         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1424                 error = xfs_iread_extents(tp, ip, whichfork);
1425                 if (error)
1426                         return error;
1427         }
1428
1429         if (xfs_iext_lookup_extent(ip, ifp, *last_block - 1, &idx, &got)) {
1430                 if (got.br_startoff <= *last_block - 1)
1431                         return 0;
1432         }
1433
1434         if (xfs_iext_get_extent(ifp, idx - 1, &got)) {
1435                 *last_block = got.br_startoff + got.br_blockcount;
1436                 return 0;
1437         }
1438
1439         *last_block = 0;
1440         return 0;
1441 }
1442
1443 int
1444 xfs_bmap_last_extent(
1445         struct xfs_trans        *tp,
1446         struct xfs_inode        *ip,
1447         int                     whichfork,
1448         struct xfs_bmbt_irec    *rec,
1449         int                     *is_empty)
1450 {
1451         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1452         int                     error;
1453         int                     nextents;
1454
1455         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1456                 error = xfs_iread_extents(tp, ip, whichfork);
1457                 if (error)
1458                         return error;
1459         }
1460
1461         nextents = xfs_iext_count(ifp);
1462         if (nextents == 0) {
1463                 *is_empty = 1;
1464                 return 0;
1465         }
1466
1467         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1468         *is_empty = 0;
1469         return 0;
1470 }
1471
1472 /*
1473  * Check the last inode extent to determine whether this allocation will result
1474  * in blocks being allocated at the end of the file. When we allocate new data
1475  * blocks at the end of the file which do not start at the previous data block,
1476  * we will try to align the new blocks at stripe unit boundaries.
1477  *
1478  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1479  * at, or past the EOF.
1480  */
1481 STATIC int
1482 xfs_bmap_isaeof(
1483         struct xfs_bmalloca     *bma,
1484         int                     whichfork)
1485 {
1486         struct xfs_bmbt_irec    rec;
1487         int                     is_empty;
1488         int                     error;
1489
1490         bma->aeof = 0;
1491         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1492                                      &is_empty);
1493         if (error)
1494                 return error;
1495
1496         if (is_empty) {
1497                 bma->aeof = 1;
1498                 return 0;
1499         }
1500
1501         /*
1502          * Check if we are allocation or past the last extent, or at least into
1503          * the last delayed allocated extent.
1504          */
1505         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1506                 (bma->offset >= rec.br_startoff &&
1507                  isnullstartblock(rec.br_startblock));
1508         return 0;
1509 }
1510
1511 /*
1512  * Returns the file-relative block number of the first block past eof in
1513  * the file.  This is not based on i_size, it is based on the extent records.
1514  * Returns 0 for local files, as they do not have extent records.
1515  */
1516 int
1517 xfs_bmap_last_offset(
1518         struct xfs_inode        *ip,
1519         xfs_fileoff_t           *last_block,
1520         int                     whichfork)
1521 {
1522         struct xfs_bmbt_irec    rec;
1523         int                     is_empty;
1524         int                     error;
1525
1526         *last_block = 0;
1527
1528         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1529                 return 0;
1530
1531         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1532             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1533                return -EIO;
1534
1535         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1536         if (error || is_empty)
1537                 return error;
1538
1539         *last_block = rec.br_startoff + rec.br_blockcount;
1540         return 0;
1541 }
1542
1543 /*
1544  * Returns whether the selected fork of the inode has exactly one
1545  * block or not.  For the data fork we check this matches di_size,
1546  * implying the file's range is 0..bsize-1.
1547  */
1548 int                                     /* 1=>1 block, 0=>otherwise */
1549 xfs_bmap_one_block(
1550         xfs_inode_t     *ip,            /* incore inode */
1551         int             whichfork)      /* data or attr fork */
1552 {
1553         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
1554         xfs_ifork_t     *ifp;           /* inode fork pointer */
1555         int             rval;           /* return value */
1556         xfs_bmbt_irec_t s;              /* internal version of extent */
1557
1558 #ifndef DEBUG
1559         if (whichfork == XFS_DATA_FORK)
1560                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1561 #endif  /* !DEBUG */
1562         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1563                 return 0;
1564         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1565                 return 0;
1566         ifp = XFS_IFORK_PTR(ip, whichfork);
1567         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1568         ep = xfs_iext_get_ext(ifp, 0);
1569         xfs_bmbt_get_all(ep, &s);
1570         rval = s.br_startoff == 0 && s.br_blockcount == 1;
1571         if (rval && whichfork == XFS_DATA_FORK)
1572                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1573         return rval;
1574 }
1575
1576 /*
1577  * Extent tree manipulation functions used during allocation.
1578  */
1579
1580 /*
1581  * Convert a delayed allocation to a real allocation.
1582  */
1583 STATIC int                              /* error */
1584 xfs_bmap_add_extent_delay_real(
1585         struct xfs_bmalloca     *bma,
1586         int                     whichfork)
1587 {
1588         struct xfs_bmbt_irec    *new = &bma->got;
1589         int                     diff;   /* temp value */
1590         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1591         int                     error;  /* error return value */
1592         int                     i;      /* temp state */
1593         xfs_ifork_t             *ifp;   /* inode fork pointer */
1594         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1595         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1596                                         /* left is 0, right is 1, prev is 2 */
1597         int                     rval=0; /* return value (logging flags) */
1598         int                     state = 0;/* state bits, accessed thru macros */
1599         xfs_filblks_t           da_new; /* new count del alloc blocks used */
1600         xfs_filblks_t           da_old; /* old count del alloc blocks used */
1601         xfs_filblks_t           temp=0; /* value for da_new calculations */
1602         xfs_filblks_t           temp2=0;/* value for da_new calculations */
1603         int                     tmp_rval;       /* partial logging flags */
1604         struct xfs_mount        *mp;
1605         xfs_extnum_t            *nextents;
1606
1607         mp = bma->ip->i_mount;
1608         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1609         ASSERT(whichfork != XFS_ATTR_FORK);
1610         nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents :
1611                                                 &bma->ip->i_d.di_nextents);
1612
1613         ASSERT(bma->idx >= 0);
1614         ASSERT(bma->idx <= xfs_iext_count(ifp));
1615         ASSERT(!isnullstartblock(new->br_startblock));
1616         ASSERT(!bma->cur ||
1617                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1618
1619         XFS_STATS_INC(mp, xs_add_exlist);
1620
1621 #define LEFT            r[0]
1622 #define RIGHT           r[1]
1623 #define PREV            r[2]
1624
1625         if (whichfork == XFS_COW_FORK)
1626                 state |= BMAP_COWFORK;
1627
1628         /*
1629          * Set up a bunch of variables to make the tests simpler.
1630          */
1631         ep = xfs_iext_get_ext(ifp, bma->idx);
1632         xfs_bmbt_get_all(ep, &PREV);
1633         new_endoff = new->br_startoff + new->br_blockcount;
1634         ASSERT(PREV.br_startoff <= new->br_startoff);
1635         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1636
1637         da_old = startblockval(PREV.br_startblock);
1638         da_new = 0;
1639
1640         /*
1641          * Set flags determining what part of the previous delayed allocation
1642          * extent is being replaced by a real allocation.
1643          */
1644         if (PREV.br_startoff == new->br_startoff)
1645                 state |= BMAP_LEFT_FILLING;
1646         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1647                 state |= BMAP_RIGHT_FILLING;
1648
1649         /*
1650          * Check and set flags if this segment has a left neighbor.
1651          * Don't set contiguous if the combined extent would be too large.
1652          */
1653         if (bma->idx > 0) {
1654                 state |= BMAP_LEFT_VALID;
1655                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1656
1657                 if (isnullstartblock(LEFT.br_startblock))
1658                         state |= BMAP_LEFT_DELAY;
1659         }
1660
1661         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1662             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1663             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1664             LEFT.br_state == new->br_state &&
1665             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1666                 state |= BMAP_LEFT_CONTIG;
1667
1668         /*
1669          * Check and set flags if this segment has a right neighbor.
1670          * Don't set contiguous if the combined extent would be too large.
1671          * Also check for all-three-contiguous being too large.
1672          */
1673         if (bma->idx < xfs_iext_count(ifp) - 1) {
1674                 state |= BMAP_RIGHT_VALID;
1675                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1676
1677                 if (isnullstartblock(RIGHT.br_startblock))
1678                         state |= BMAP_RIGHT_DELAY;
1679         }
1680
1681         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1682             new_endoff == RIGHT.br_startoff &&
1683             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1684             new->br_state == RIGHT.br_state &&
1685             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1686             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1687                        BMAP_RIGHT_FILLING)) !=
1688                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1689                        BMAP_RIGHT_FILLING) ||
1690              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1691                         <= MAXEXTLEN))
1692                 state |= BMAP_RIGHT_CONTIG;
1693
1694         error = 0;
1695         /*
1696          * Switch out based on the FILLING and CONTIG state bits.
1697          */
1698         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1699                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1700         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1701              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1702                 /*
1703                  * Filling in all of a previously delayed allocation extent.
1704                  * The left and right neighbors are both contiguous with new.
1705                  */
1706                 bma->idx--;
1707                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1708                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1709                         LEFT.br_blockcount + PREV.br_blockcount +
1710                         RIGHT.br_blockcount);
1711                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1712
1713                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1714                 (*nextents)--;
1715                 if (bma->cur == NULL)
1716                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1717                 else {
1718                         rval = XFS_ILOG_CORE;
1719                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1720                                         RIGHT.br_startblock,
1721                                         RIGHT.br_blockcount, &i);
1722                         if (error)
1723                                 goto done;
1724                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1725                         error = xfs_btree_delete(bma->cur, &i);
1726                         if (error)
1727                                 goto done;
1728                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1729                         error = xfs_btree_decrement(bma->cur, 0, &i);
1730                         if (error)
1731                                 goto done;
1732                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1733                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1734                                         LEFT.br_startblock,
1735                                         LEFT.br_blockcount +
1736                                         PREV.br_blockcount +
1737                                         RIGHT.br_blockcount, LEFT.br_state);
1738                         if (error)
1739                                 goto done;
1740                 }
1741                 break;
1742
1743         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1744                 /*
1745                  * Filling in all of a previously delayed allocation extent.
1746                  * The left neighbor is contiguous, the right is not.
1747                  */
1748                 bma->idx--;
1749
1750                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1751                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1752                         LEFT.br_blockcount + PREV.br_blockcount);
1753                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1754
1755                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1756                 if (bma->cur == NULL)
1757                         rval = XFS_ILOG_DEXT;
1758                 else {
1759                         rval = 0;
1760                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1761                                         LEFT.br_startblock, LEFT.br_blockcount,
1762                                         &i);
1763                         if (error)
1764                                 goto done;
1765                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1766                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1767                                         LEFT.br_startblock,
1768                                         LEFT.br_blockcount +
1769                                         PREV.br_blockcount, LEFT.br_state);
1770                         if (error)
1771                                 goto done;
1772                 }
1773                 break;
1774
1775         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1776                 /*
1777                  * Filling in all of a previously delayed allocation extent.
1778                  * The right neighbor is contiguous, the left is not.
1779                  */
1780                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1781                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1782                 xfs_bmbt_set_blockcount(ep,
1783                         PREV.br_blockcount + RIGHT.br_blockcount);
1784                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1785
1786                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1787                 if (bma->cur == NULL)
1788                         rval = XFS_ILOG_DEXT;
1789                 else {
1790                         rval = 0;
1791                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1792                                         RIGHT.br_startblock,
1793                                         RIGHT.br_blockcount, &i);
1794                         if (error)
1795                                 goto done;
1796                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1797                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1798                                         new->br_startblock,
1799                                         PREV.br_blockcount +
1800                                         RIGHT.br_blockcount, PREV.br_state);
1801                         if (error)
1802                                 goto done;
1803                 }
1804                 break;
1805
1806         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1807                 /*
1808                  * Filling in all of a previously delayed allocation extent.
1809                  * Neither the left nor right neighbors are contiguous with
1810                  * the new one.
1811                  */
1812                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1813                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1814                 xfs_bmbt_set_state(ep, new->br_state);
1815                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1816
1817                 (*nextents)++;
1818                 if (bma->cur == NULL)
1819                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1820                 else {
1821                         rval = XFS_ILOG_CORE;
1822                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1823                                         new->br_startblock, new->br_blockcount,
1824                                         &i);
1825                         if (error)
1826                                 goto done;
1827                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1828                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1829                         error = xfs_btree_insert(bma->cur, &i);
1830                         if (error)
1831                                 goto done;
1832                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1833                 }
1834                 break;
1835
1836         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1837                 /*
1838                  * Filling in the first part of a previous delayed allocation.
1839                  * The left neighbor is contiguous.
1840                  */
1841                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1842                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1843                         LEFT.br_blockcount + new->br_blockcount);
1844                 xfs_bmbt_set_startoff(ep,
1845                         PREV.br_startoff + new->br_blockcount);
1846                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1847
1848                 temp = PREV.br_blockcount - new->br_blockcount;
1849                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1850                 xfs_bmbt_set_blockcount(ep, temp);
1851                 if (bma->cur == NULL)
1852                         rval = XFS_ILOG_DEXT;
1853                 else {
1854                         rval = 0;
1855                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1856                                         LEFT.br_startblock, LEFT.br_blockcount,
1857                                         &i);
1858                         if (error)
1859                                 goto done;
1860                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1861                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1862                                         LEFT.br_startblock,
1863                                         LEFT.br_blockcount +
1864                                         new->br_blockcount,
1865                                         LEFT.br_state);
1866                         if (error)
1867                                 goto done;
1868                 }
1869                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1870                         startblockval(PREV.br_startblock));
1871                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1872                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1873
1874                 bma->idx--;
1875                 break;
1876
1877         case BMAP_LEFT_FILLING:
1878                 /*
1879                  * Filling in the first part of a previous delayed allocation.
1880                  * The left neighbor is not contiguous.
1881                  */
1882                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1883                 xfs_bmbt_set_startoff(ep, new_endoff);
1884                 temp = PREV.br_blockcount - new->br_blockcount;
1885                 xfs_bmbt_set_blockcount(ep, temp);
1886                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1887                 (*nextents)++;
1888                 if (bma->cur == NULL)
1889                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1890                 else {
1891                         rval = XFS_ILOG_CORE;
1892                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1893                                         new->br_startblock, new->br_blockcount,
1894                                         &i);
1895                         if (error)
1896                                 goto done;
1897                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1898                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1899                         error = xfs_btree_insert(bma->cur, &i);
1900                         if (error)
1901                                 goto done;
1902                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1903                 }
1904
1905                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1906                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1907                                         bma->firstblock, bma->dfops,
1908                                         &bma->cur, 1, &tmp_rval, whichfork);
1909                         rval |= tmp_rval;
1910                         if (error)
1911                                 goto done;
1912                 }
1913                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1914                         startblockval(PREV.br_startblock) -
1915                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
1916                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
1917                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1918                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1919                 break;
1920
1921         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1922                 /*
1923                  * Filling in the last part of a previous delayed allocation.
1924                  * The right neighbor is contiguous with the new allocation.
1925                  */
1926                 temp = PREV.br_blockcount - new->br_blockcount;
1927                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1928                 xfs_bmbt_set_blockcount(ep, temp);
1929                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
1930                         new->br_startoff, new->br_startblock,
1931                         new->br_blockcount + RIGHT.br_blockcount,
1932                         RIGHT.br_state);
1933                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1934                 if (bma->cur == NULL)
1935                         rval = XFS_ILOG_DEXT;
1936                 else {
1937                         rval = 0;
1938                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1939                                         RIGHT.br_startblock,
1940                                         RIGHT.br_blockcount, &i);
1941                         if (error)
1942                                 goto done;
1943                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1944                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
1945                                         new->br_startblock,
1946                                         new->br_blockcount +
1947                                         RIGHT.br_blockcount,
1948                                         RIGHT.br_state);
1949                         if (error)
1950                                 goto done;
1951                 }
1952
1953                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1954                         startblockval(PREV.br_startblock));
1955                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1956                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1957                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1958
1959                 bma->idx++;
1960                 break;
1961
1962         case BMAP_RIGHT_FILLING:
1963                 /*
1964                  * Filling in the last part of a previous delayed allocation.
1965                  * The right neighbor is not contiguous.
1966                  */
1967                 temp = PREV.br_blockcount - new->br_blockcount;
1968                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1969                 xfs_bmbt_set_blockcount(ep, temp);
1970                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
1971                 (*nextents)++;
1972                 if (bma->cur == NULL)
1973                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1974                 else {
1975                         rval = XFS_ILOG_CORE;
1976                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1977                                         new->br_startblock, new->br_blockcount,
1978                                         &i);
1979                         if (error)
1980                                 goto done;
1981                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1982                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1983                         error = xfs_btree_insert(bma->cur, &i);
1984                         if (error)
1985                                 goto done;
1986                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1987                 }
1988
1989                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1990                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1991                                 bma->firstblock, bma->dfops, &bma->cur, 1,
1992                                 &tmp_rval, whichfork);
1993                         rval |= tmp_rval;
1994                         if (error)
1995                                 goto done;
1996                 }
1997                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1998                         startblockval(PREV.br_startblock) -
1999                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2000                 ep = xfs_iext_get_ext(ifp, bma->idx);
2001                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2002                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2003
2004                 bma->idx++;
2005                 break;
2006
2007         case 0:
2008                 /*
2009                  * Filling in the middle part of a previous delayed allocation.
2010                  * Contiguity is impossible here.
2011                  * This case is avoided almost all the time.
2012                  *
2013                  * We start with a delayed allocation:
2014                  *
2015                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2016                  *  PREV @ idx
2017                  *
2018                  * and we are allocating:
2019                  *                     +rrrrrrrrrrrrrrrrr+
2020                  *                            new
2021                  *
2022                  * and we set it up for insertion as:
2023                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2024                  *                            new
2025                  *  PREV @ idx          LEFT              RIGHT
2026                  *                      inserted at idx + 1
2027                  */
2028                 temp = new->br_startoff - PREV.br_startoff;
2029                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2030                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2031                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
2032                 LEFT = *new;
2033                 RIGHT.br_state = PREV.br_state;
2034                 RIGHT.br_startblock = nullstartblock(
2035                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2036                 RIGHT.br_startoff = new_endoff;
2037                 RIGHT.br_blockcount = temp2;
2038                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2039                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2040                 (*nextents)++;
2041                 if (bma->cur == NULL)
2042                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2043                 else {
2044                         rval = XFS_ILOG_CORE;
2045                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2046                                         new->br_startblock, new->br_blockcount,
2047                                         &i);
2048                         if (error)
2049                                 goto done;
2050                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2051                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2052                         error = xfs_btree_insert(bma->cur, &i);
2053                         if (error)
2054                                 goto done;
2055                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2056                 }
2057
2058                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2059                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2060                                         bma->firstblock, bma->dfops, &bma->cur,
2061                                         1, &tmp_rval, whichfork);
2062                         rval |= tmp_rval;
2063                         if (error)
2064                                 goto done;
2065                 }
2066                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2067                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2068                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2069                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2070                 if (diff > 0) {
2071                         error = xfs_mod_fdblocks(bma->ip->i_mount,
2072                                                  -((int64_t)diff), false);
2073                         ASSERT(!error);
2074                         if (error)
2075                                 goto done;
2076                 }
2077
2078                 ep = xfs_iext_get_ext(ifp, bma->idx);
2079                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2080                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2081                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2082                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2083                         nullstartblock((int)temp2));
2084                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2085
2086                 bma->idx++;
2087                 da_new = temp + temp2;
2088                 break;
2089
2090         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2091         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2092         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2093         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2094         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2095         case BMAP_LEFT_CONTIG:
2096         case BMAP_RIGHT_CONTIG:
2097                 /*
2098                  * These cases are all impossible.
2099                  */
2100                 ASSERT(0);
2101         }
2102
2103         /* add reverse mapping */
2104         error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new);
2105         if (error)
2106                 goto done;
2107
2108         /* convert to a btree if necessary */
2109         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2110                 int     tmp_logflags;   /* partial log flag return val */
2111
2112                 ASSERT(bma->cur == NULL);
2113                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2114                                 bma->firstblock, bma->dfops, &bma->cur,
2115                                 da_old > 0, &tmp_logflags, whichfork);
2116                 bma->logflags |= tmp_logflags;
2117                 if (error)
2118                         goto done;
2119         }
2120
2121         /* adjust for changes in reserved delayed indirect blocks */
2122         if (da_old || da_new) {
2123                 temp = da_new;
2124                 if (bma->cur)
2125                         temp += bma->cur->bc_private.b.allocated;
2126                 ASSERT(temp <= da_old);
2127                 if (temp < da_old)
2128                         xfs_mod_fdblocks(bma->ip->i_mount,
2129                                         (int64_t)(da_old - temp), false);
2130         }
2131
2132         /* clear out the allocated field, done with it now in any case. */
2133         if (bma->cur)
2134                 bma->cur->bc_private.b.allocated = 0;
2135
2136         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2137 done:
2138         if (whichfork != XFS_COW_FORK)
2139                 bma->logflags |= rval;
2140         return error;
2141 #undef  LEFT
2142 #undef  RIGHT
2143 #undef  PREV
2144 }
2145
2146 /*
2147  * Convert an unwritten allocation to a real allocation or vice versa.
2148  */
2149 STATIC int                              /* error */
2150 xfs_bmap_add_extent_unwritten_real(
2151         struct xfs_trans        *tp,
2152         xfs_inode_t             *ip,    /* incore inode pointer */
2153         int                     whichfork,
2154         xfs_extnum_t            *idx,   /* extent number to update/insert */
2155         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2156         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2157         xfs_fsblock_t           *first, /* pointer to firstblock variable */
2158         struct xfs_defer_ops    *dfops, /* list of extents to be freed */
2159         int                     *logflagsp) /* inode logging flags */
2160 {
2161         xfs_btree_cur_t         *cur;   /* btree cursor */
2162         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2163         int                     error;  /* error return value */
2164         int                     i;      /* temp state */
2165         xfs_ifork_t             *ifp;   /* inode fork pointer */
2166         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2167         xfs_exntst_t            newext; /* new extent state */
2168         xfs_exntst_t            oldext; /* old extent state */
2169         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2170                                         /* left is 0, right is 1, prev is 2 */
2171         int                     rval=0; /* return value (logging flags) */
2172         int                     state = 0;/* state bits, accessed thru macros */
2173         struct xfs_mount        *mp = ip->i_mount;
2174
2175         *logflagsp = 0;
2176
2177         cur = *curp;
2178         ifp = XFS_IFORK_PTR(ip, whichfork);
2179         if (whichfork == XFS_COW_FORK)
2180                 state |= BMAP_COWFORK;
2181
2182         ASSERT(*idx >= 0);
2183         ASSERT(*idx <= xfs_iext_count(ifp));
2184         ASSERT(!isnullstartblock(new->br_startblock));
2185
2186         XFS_STATS_INC(mp, xs_add_exlist);
2187
2188 #define LEFT            r[0]
2189 #define RIGHT           r[1]
2190 #define PREV            r[2]
2191
2192         /*
2193          * Set up a bunch of variables to make the tests simpler.
2194          */
2195         error = 0;
2196         ep = xfs_iext_get_ext(ifp, *idx);
2197         xfs_bmbt_get_all(ep, &PREV);
2198         newext = new->br_state;
2199         oldext = (newext == XFS_EXT_UNWRITTEN) ?
2200                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2201         ASSERT(PREV.br_state == oldext);
2202         new_endoff = new->br_startoff + new->br_blockcount;
2203         ASSERT(PREV.br_startoff <= new->br_startoff);
2204         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2205
2206         /*
2207          * Set flags determining what part of the previous oldext allocation
2208          * extent is being replaced by a newext allocation.
2209          */
2210         if (PREV.br_startoff == new->br_startoff)
2211                 state |= BMAP_LEFT_FILLING;
2212         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2213                 state |= BMAP_RIGHT_FILLING;
2214
2215         /*
2216          * Check and set flags if this segment has a left neighbor.
2217          * Don't set contiguous if the combined extent would be too large.
2218          */
2219         if (*idx > 0) {
2220                 state |= BMAP_LEFT_VALID;
2221                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2222
2223                 if (isnullstartblock(LEFT.br_startblock))
2224                         state |= BMAP_LEFT_DELAY;
2225         }
2226
2227         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2228             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2229             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2230             LEFT.br_state == newext &&
2231             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2232                 state |= BMAP_LEFT_CONTIG;
2233
2234         /*
2235          * Check and set flags if this segment has a right neighbor.
2236          * Don't set contiguous if the combined extent would be too large.
2237          * Also check for all-three-contiguous being too large.
2238          */
2239         if (*idx < xfs_iext_count(ifp) - 1) {
2240                 state |= BMAP_RIGHT_VALID;
2241                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2242                 if (isnullstartblock(RIGHT.br_startblock))
2243                         state |= BMAP_RIGHT_DELAY;
2244         }
2245
2246         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2247             new_endoff == RIGHT.br_startoff &&
2248             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2249             newext == RIGHT.br_state &&
2250             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2251             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2252                        BMAP_RIGHT_FILLING)) !=
2253                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2254                        BMAP_RIGHT_FILLING) ||
2255              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2256                         <= MAXEXTLEN))
2257                 state |= BMAP_RIGHT_CONTIG;
2258
2259         /*
2260          * Switch out based on the FILLING and CONTIG state bits.
2261          */
2262         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2263                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2264         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2265              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2266                 /*
2267                  * Setting all of a previous oldext extent to newext.
2268                  * The left and right neighbors are both contiguous with new.
2269                  */
2270                 --*idx;
2271
2272                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2273                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2274                         LEFT.br_blockcount + PREV.br_blockcount +
2275                         RIGHT.br_blockcount);
2276                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2277
2278                 xfs_iext_remove(ip, *idx + 1, 2, state);
2279                 XFS_IFORK_NEXT_SET(ip, whichfork,
2280                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 2);
2281                 if (cur == NULL)
2282                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2283                 else {
2284                         rval = XFS_ILOG_CORE;
2285                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2286                                         RIGHT.br_startblock,
2287                                         RIGHT.br_blockcount, &i)))
2288                                 goto done;
2289                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2290                         if ((error = xfs_btree_delete(cur, &i)))
2291                                 goto done;
2292                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2293                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2294                                 goto done;
2295                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2296                         if ((error = xfs_btree_delete(cur, &i)))
2297                                 goto done;
2298                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2299                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2300                                 goto done;
2301                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2302                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2303                                 LEFT.br_startblock,
2304                                 LEFT.br_blockcount + PREV.br_blockcount +
2305                                 RIGHT.br_blockcount, LEFT.br_state)))
2306                                 goto done;
2307                 }
2308                 break;
2309
2310         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2311                 /*
2312                  * Setting all of a previous oldext extent to newext.
2313                  * The left neighbor is contiguous, the right is not.
2314                  */
2315                 --*idx;
2316
2317                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2318                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2319                         LEFT.br_blockcount + PREV.br_blockcount);
2320                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2321
2322                 xfs_iext_remove(ip, *idx + 1, 1, state);
2323                 XFS_IFORK_NEXT_SET(ip, whichfork,
2324                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2325                 if (cur == NULL)
2326                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2327                 else {
2328                         rval = XFS_ILOG_CORE;
2329                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2330                                         PREV.br_startblock, PREV.br_blockcount,
2331                                         &i)))
2332                                 goto done;
2333                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2334                         if ((error = xfs_btree_delete(cur, &i)))
2335                                 goto done;
2336                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2337                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2338                                 goto done;
2339                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2340                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2341                                 LEFT.br_startblock,
2342                                 LEFT.br_blockcount + PREV.br_blockcount,
2343                                 LEFT.br_state)))
2344                                 goto done;
2345                 }
2346                 break;
2347
2348         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2349                 /*
2350                  * Setting all of a previous oldext extent to newext.
2351                  * The right neighbor is contiguous, the left is not.
2352                  */
2353                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2354                 xfs_bmbt_set_blockcount(ep,
2355                         PREV.br_blockcount + RIGHT.br_blockcount);
2356                 xfs_bmbt_set_state(ep, newext);
2357                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2358                 xfs_iext_remove(ip, *idx + 1, 1, state);
2359                 XFS_IFORK_NEXT_SET(ip, whichfork,
2360                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2361                 if (cur == NULL)
2362                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2363                 else {
2364                         rval = XFS_ILOG_CORE;
2365                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2366                                         RIGHT.br_startblock,
2367                                         RIGHT.br_blockcount, &i)))
2368                                 goto done;
2369                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2370                         if ((error = xfs_btree_delete(cur, &i)))
2371                                 goto done;
2372                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2373                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2374                                 goto done;
2375                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2376                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2377                                 new->br_startblock,
2378                                 new->br_blockcount + RIGHT.br_blockcount,
2379                                 newext)))
2380                                 goto done;
2381                 }
2382                 break;
2383
2384         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2385                 /*
2386                  * Setting all of a previous oldext extent to newext.
2387                  * Neither the left nor right neighbors are contiguous with
2388                  * the new one.
2389                  */
2390                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2391                 xfs_bmbt_set_state(ep, newext);
2392                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2393
2394                 if (cur == NULL)
2395                         rval = XFS_ILOG_DEXT;
2396                 else {
2397                         rval = 0;
2398                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2399                                         new->br_startblock, new->br_blockcount,
2400                                         &i)))
2401                                 goto done;
2402                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2403                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2404                                 new->br_startblock, new->br_blockcount,
2405                                 newext)))
2406                                 goto done;
2407                 }
2408                 break;
2409
2410         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2411                 /*
2412                  * Setting the first part of a previous oldext extent to newext.
2413                  * The left neighbor is contiguous.
2414                  */
2415                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2416                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2417                         LEFT.br_blockcount + new->br_blockcount);
2418                 xfs_bmbt_set_startoff(ep,
2419                         PREV.br_startoff + new->br_blockcount);
2420                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2421
2422                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2423                 xfs_bmbt_set_startblock(ep,
2424                         new->br_startblock + new->br_blockcount);
2425                 xfs_bmbt_set_blockcount(ep,
2426                         PREV.br_blockcount - new->br_blockcount);
2427                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2428
2429                 --*idx;
2430
2431                 if (cur == NULL)
2432                         rval = XFS_ILOG_DEXT;
2433                 else {
2434                         rval = 0;
2435                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2436                                         PREV.br_startblock, PREV.br_blockcount,
2437                                         &i)))
2438                                 goto done;
2439                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2440                         if ((error = xfs_bmbt_update(cur,
2441                                 PREV.br_startoff + new->br_blockcount,
2442                                 PREV.br_startblock + new->br_blockcount,
2443                                 PREV.br_blockcount - new->br_blockcount,
2444                                 oldext)))
2445                                 goto done;
2446                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2447                                 goto done;
2448                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
2449                                 LEFT.br_startblock,
2450                                 LEFT.br_blockcount + new->br_blockcount,
2451                                 LEFT.br_state);
2452                         if (error)
2453                                 goto done;
2454                 }
2455                 break;
2456
2457         case BMAP_LEFT_FILLING:
2458                 /*
2459                  * Setting the first part of a previous oldext extent to newext.
2460                  * The left neighbor is not contiguous.
2461                  */
2462                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2463                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2464                 xfs_bmbt_set_startoff(ep, new_endoff);
2465                 xfs_bmbt_set_blockcount(ep,
2466                         PREV.br_blockcount - new->br_blockcount);
2467                 xfs_bmbt_set_startblock(ep,
2468                         new->br_startblock + new->br_blockcount);
2469                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2470
2471                 xfs_iext_insert(ip, *idx, 1, new, state);
2472                 XFS_IFORK_NEXT_SET(ip, whichfork,
2473                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2474                 if (cur == NULL)
2475                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2476                 else {
2477                         rval = XFS_ILOG_CORE;
2478                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2479                                         PREV.br_startblock, PREV.br_blockcount,
2480                                         &i)))
2481                                 goto done;
2482                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2483                         if ((error = xfs_bmbt_update(cur,
2484                                 PREV.br_startoff + new->br_blockcount,
2485                                 PREV.br_startblock + new->br_blockcount,
2486                                 PREV.br_blockcount - new->br_blockcount,
2487                                 oldext)))
2488                                 goto done;
2489                         cur->bc_rec.b = *new;
2490                         if ((error = xfs_btree_insert(cur, &i)))
2491                                 goto done;
2492                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2493                 }
2494                 break;
2495
2496         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2497                 /*
2498                  * Setting the last part of a previous oldext extent to newext.
2499                  * The right neighbor is contiguous with the new allocation.
2500                  */
2501                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2502                 xfs_bmbt_set_blockcount(ep,
2503                         PREV.br_blockcount - new->br_blockcount);
2504                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2505
2506                 ++*idx;
2507
2508                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2509                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2510                         new->br_startoff, new->br_startblock,
2511                         new->br_blockcount + RIGHT.br_blockcount, newext);
2512                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2513
2514                 if (cur == NULL)
2515                         rval = XFS_ILOG_DEXT;
2516                 else {
2517                         rval = 0;
2518                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2519                                         PREV.br_startblock,
2520                                         PREV.br_blockcount, &i)))
2521                                 goto done;
2522                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2523                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2524                                 PREV.br_startblock,
2525                                 PREV.br_blockcount - new->br_blockcount,
2526                                 oldext)))
2527                                 goto done;
2528                         if ((error = xfs_btree_increment(cur, 0, &i)))
2529                                 goto done;
2530                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2531                                 new->br_startblock,
2532                                 new->br_blockcount + RIGHT.br_blockcount,
2533                                 newext)))
2534                                 goto done;
2535                 }
2536                 break;
2537
2538         case BMAP_RIGHT_FILLING:
2539                 /*
2540                  * Setting the last part of a previous oldext extent to newext.
2541                  * The right neighbor is not contiguous.
2542                  */
2543                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2544                 xfs_bmbt_set_blockcount(ep,
2545                         PREV.br_blockcount - new->br_blockcount);
2546                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2547
2548                 ++*idx;
2549                 xfs_iext_insert(ip, *idx, 1, new, state);
2550
2551                 XFS_IFORK_NEXT_SET(ip, whichfork,
2552                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2553                 if (cur == NULL)
2554                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2555                 else {
2556                         rval = XFS_ILOG_CORE;
2557                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2558                                         PREV.br_startblock, PREV.br_blockcount,
2559                                         &i)))
2560                                 goto done;
2561                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2562                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2563                                 PREV.br_startblock,
2564                                 PREV.br_blockcount - new->br_blockcount,
2565                                 oldext)))
2566                                 goto done;
2567                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2568                                         new->br_startblock, new->br_blockcount,
2569                                         &i)))
2570                                 goto done;
2571                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2572                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
2573                         if ((error = xfs_btree_insert(cur, &i)))
2574                                 goto done;
2575                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2576                 }
2577                 break;
2578
2579         case 0:
2580                 /*
2581                  * Setting the middle part of a previous oldext extent to
2582                  * newext.  Contiguity is impossible here.
2583                  * One extent becomes three extents.
2584                  */
2585                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2586                 xfs_bmbt_set_blockcount(ep,
2587                         new->br_startoff - PREV.br_startoff);
2588                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2589
2590                 r[0] = *new;
2591                 r[1].br_startoff = new_endoff;
2592                 r[1].br_blockcount =
2593                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
2594                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2595                 r[1].br_state = oldext;
2596
2597                 ++*idx;
2598                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2599
2600                 XFS_IFORK_NEXT_SET(ip, whichfork,
2601                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 2);
2602                 if (cur == NULL)
2603                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2604                 else {
2605                         rval = XFS_ILOG_CORE;
2606                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2607                                         PREV.br_startblock, PREV.br_blockcount,
2608                                         &i)))
2609                                 goto done;
2610                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2611                         /* new right extent - oldext */
2612                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2613                                 r[1].br_startblock, r[1].br_blockcount,
2614                                 r[1].br_state)))
2615                                 goto done;
2616                         /* new left extent - oldext */
2617                         cur->bc_rec.b = PREV;
2618                         cur->bc_rec.b.br_blockcount =
2619                                 new->br_startoff - PREV.br_startoff;
2620                         if ((error = xfs_btree_insert(cur, &i)))
2621                                 goto done;
2622                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2623                         /*
2624                          * Reset the cursor to the position of the new extent
2625                          * we are about to insert as we can't trust it after
2626                          * the previous insert.
2627                          */
2628                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2629                                         new->br_startblock, new->br_blockcount,
2630                                         &i)))
2631                                 goto done;
2632                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2633                         /* new middle extent - newext */
2634                         cur->bc_rec.b.br_state = new->br_state;
2635                         if ((error = xfs_btree_insert(cur, &i)))
2636                                 goto done;
2637                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2638                 }
2639                 break;
2640
2641         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2642         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2643         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2644         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2645         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2646         case BMAP_LEFT_CONTIG:
2647         case BMAP_RIGHT_CONTIG:
2648                 /*
2649                  * These cases are all impossible.
2650                  */
2651                 ASSERT(0);
2652         }
2653
2654         /* update reverse mappings */
2655         error = xfs_rmap_convert_extent(mp, dfops, ip, whichfork, new);
2656         if (error)
2657                 goto done;
2658
2659         /* convert to a btree if necessary */
2660         if (xfs_bmap_needs_btree(ip, whichfork)) {
2661                 int     tmp_logflags;   /* partial log flag return val */
2662
2663                 ASSERT(cur == NULL);
2664                 error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, &cur,
2665                                 0, &tmp_logflags, whichfork);
2666                 *logflagsp |= tmp_logflags;
2667                 if (error)
2668                         goto done;
2669         }
2670
2671         /* clear out the allocated field, done with it now in any case. */
2672         if (cur) {
2673                 cur->bc_private.b.allocated = 0;
2674                 *curp = cur;
2675         }
2676
2677         xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2678 done:
2679         *logflagsp |= rval;
2680         return error;
2681 #undef  LEFT
2682 #undef  RIGHT
2683 #undef  PREV
2684 }
2685
2686 /*
2687  * Convert a hole to a delayed allocation.
2688  */
2689 STATIC void
2690 xfs_bmap_add_extent_hole_delay(
2691         xfs_inode_t             *ip,    /* incore inode pointer */
2692         int                     whichfork,
2693         xfs_extnum_t            *idx,   /* extent number to update/insert */
2694         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2695 {
2696         xfs_ifork_t             *ifp;   /* inode fork pointer */
2697         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2698         xfs_filblks_t           newlen=0;       /* new indirect size */
2699         xfs_filblks_t           oldlen=0;       /* old indirect size */
2700         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2701         int                     state;  /* state bits, accessed thru macros */
2702         xfs_filblks_t           temp=0; /* temp for indirect calculations */
2703
2704         ifp = XFS_IFORK_PTR(ip, whichfork);
2705         state = 0;
2706         if (whichfork == XFS_COW_FORK)
2707                 state |= BMAP_COWFORK;
2708         ASSERT(isnullstartblock(new->br_startblock));
2709
2710         /*
2711          * Check and set flags if this segment has a left neighbor
2712          */
2713         if (*idx > 0) {
2714                 state |= BMAP_LEFT_VALID;
2715                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2716
2717                 if (isnullstartblock(left.br_startblock))
2718                         state |= BMAP_LEFT_DELAY;
2719         }
2720
2721         /*
2722          * Check and set flags if the current (right) segment exists.
2723          * If it doesn't exist, we're converting the hole at end-of-file.
2724          */
2725         if (*idx < xfs_iext_count(ifp)) {
2726                 state |= BMAP_RIGHT_VALID;
2727                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2728
2729                 if (isnullstartblock(right.br_startblock))
2730                         state |= BMAP_RIGHT_DELAY;
2731         }
2732
2733         /*
2734          * Set contiguity flags on the left and right neighbors.
2735          * Don't let extents get too large, even if the pieces are contiguous.
2736          */
2737         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2738             left.br_startoff + left.br_blockcount == new->br_startoff &&
2739             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2740                 state |= BMAP_LEFT_CONTIG;
2741
2742         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2743             new->br_startoff + new->br_blockcount == right.br_startoff &&
2744             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2745             (!(state & BMAP_LEFT_CONTIG) ||
2746              (left.br_blockcount + new->br_blockcount +
2747               right.br_blockcount <= MAXEXTLEN)))
2748                 state |= BMAP_RIGHT_CONTIG;
2749
2750         /*
2751          * Switch out based on the contiguity flags.
2752          */
2753         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2754         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2755                 /*
2756                  * New allocation is contiguous with delayed allocations
2757                  * on the left and on the right.
2758                  * Merge all three into a single extent record.
2759                  */
2760                 --*idx;
2761                 temp = left.br_blockcount + new->br_blockcount +
2762                         right.br_blockcount;
2763
2764                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2765                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2766                 oldlen = startblockval(left.br_startblock) +
2767                         startblockval(new->br_startblock) +
2768                         startblockval(right.br_startblock);
2769                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2770                                          oldlen);
2771                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2772                         nullstartblock((int)newlen));
2773                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2774
2775                 xfs_iext_remove(ip, *idx + 1, 1, state);
2776                 break;
2777
2778         case BMAP_LEFT_CONTIG:
2779                 /*
2780                  * New allocation is contiguous with a delayed allocation
2781                  * on the left.
2782                  * Merge the new allocation with the left neighbor.
2783                  */
2784                 --*idx;
2785                 temp = left.br_blockcount + new->br_blockcount;
2786
2787                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2788                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2789                 oldlen = startblockval(left.br_startblock) +
2790                         startblockval(new->br_startblock);
2791                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2792                                          oldlen);
2793                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2794                         nullstartblock((int)newlen));
2795                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2796                 break;
2797
2798         case BMAP_RIGHT_CONTIG:
2799                 /*
2800                  * New allocation is contiguous with a delayed allocation
2801                  * on the right.
2802                  * Merge the new allocation with the right neighbor.
2803                  */
2804                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2805                 temp = new->br_blockcount + right.br_blockcount;
2806                 oldlen = startblockval(new->br_startblock) +
2807                         startblockval(right.br_startblock);
2808                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2809                                          oldlen);
2810                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2811                         new->br_startoff,
2812                         nullstartblock((int)newlen), temp, right.br_state);
2813                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2814                 break;
2815
2816         case 0:
2817                 /*
2818                  * New allocation is not contiguous with another
2819                  * delayed allocation.
2820                  * Insert a new entry.
2821                  */
2822                 oldlen = newlen = 0;
2823                 xfs_iext_insert(ip, *idx, 1, new, state);
2824                 break;
2825         }
2826         if (oldlen != newlen) {
2827                 ASSERT(oldlen > newlen);
2828                 xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2829                                  false);
2830                 /*
2831                  * Nothing to do for disk quota accounting here.
2832                  */
2833         }
2834 }
2835
2836 /*
2837  * Convert a hole to a real allocation.
2838  */
2839 STATIC int                              /* error */
2840 xfs_bmap_add_extent_hole_real(
2841         struct xfs_trans        *tp,
2842         struct xfs_inode        *ip,
2843         int                     whichfork,
2844         xfs_extnum_t            *idx,
2845         struct xfs_btree_cur    **curp,
2846         struct xfs_bmbt_irec    *new,
2847         xfs_fsblock_t           *first,
2848         struct xfs_defer_ops    *dfops,
2849         int                     *logflagsp)
2850 {
2851         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
2852         struct xfs_mount        *mp = ip->i_mount;
2853         struct xfs_btree_cur    *cur = *curp;
2854         int                     error;  /* error return value */
2855         int                     i;      /* temp state */
2856         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2857         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2858         int                     rval=0; /* return value (logging flags) */
2859         int                     state;  /* state bits, accessed thru macros */
2860
2861         ASSERT(*idx >= 0);
2862         ASSERT(*idx <= xfs_iext_count(ifp));
2863         ASSERT(!isnullstartblock(new->br_startblock));
2864         ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2865
2866         XFS_STATS_INC(mp, xs_add_exlist);
2867
2868         state = 0;
2869         if (whichfork == XFS_ATTR_FORK)
2870                 state |= BMAP_ATTRFORK;
2871         if (whichfork == XFS_COW_FORK)
2872                 state |= BMAP_COWFORK;
2873
2874         /*
2875          * Check and set flags if this segment has a left neighbor.
2876          */
2877         if (*idx > 0) {
2878                 state |= BMAP_LEFT_VALID;
2879                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2880                 if (isnullstartblock(left.br_startblock))
2881                         state |= BMAP_LEFT_DELAY;
2882         }
2883
2884         /*
2885          * Check and set flags if this segment has a current value.
2886          * Not true if we're inserting into the "hole" at eof.
2887          */
2888         if (*idx < xfs_iext_count(ifp)) {
2889                 state |= BMAP_RIGHT_VALID;
2890                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2891                 if (isnullstartblock(right.br_startblock))
2892                         state |= BMAP_RIGHT_DELAY;
2893         }
2894
2895         /*
2896          * We're inserting a real allocation between "left" and "right".
2897          * Set the contiguity flags.  Don't let extents get too large.
2898          */
2899         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2900             left.br_startoff + left.br_blockcount == new->br_startoff &&
2901             left.br_startblock + left.br_blockcount == new->br_startblock &&
2902             left.br_state == new->br_state &&
2903             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2904                 state |= BMAP_LEFT_CONTIG;
2905
2906         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2907             new->br_startoff + new->br_blockcount == right.br_startoff &&
2908             new->br_startblock + new->br_blockcount == right.br_startblock &&
2909             new->br_state == right.br_state &&
2910             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2911             (!(state & BMAP_LEFT_CONTIG) ||
2912              left.br_blockcount + new->br_blockcount +
2913              right.br_blockcount <= MAXEXTLEN))
2914                 state |= BMAP_RIGHT_CONTIG;
2915
2916         error = 0;
2917         /*
2918          * Select which case we're in here, and implement it.
2919          */
2920         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2921         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2922                 /*
2923                  * New allocation is contiguous with real allocations on the
2924                  * left and on the right.
2925                  * Merge all three into a single extent record.
2926                  */
2927                 --*idx;
2928                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2929                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2930                         left.br_blockcount + new->br_blockcount +
2931                         right.br_blockcount);
2932                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2933
2934                 xfs_iext_remove(ip, *idx + 1, 1, state);
2935
2936                 XFS_IFORK_NEXT_SET(ip, whichfork,
2937                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2938                 if (cur == NULL) {
2939                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2940                 } else {
2941                         rval = XFS_ILOG_CORE;
2942                         error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2943                                         right.br_startblock, right.br_blockcount,
2944                                         &i);
2945                         if (error)
2946                                 goto done;
2947                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2948                         error = xfs_btree_delete(cur, &i);
2949                         if (error)
2950                                 goto done;
2951                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2952                         error = xfs_btree_decrement(cur, 0, &i);
2953                         if (error)
2954                                 goto done;
2955                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2956                         error = xfs_bmbt_update(cur, left.br_startoff,
2957                                         left.br_startblock,
2958                                         left.br_blockcount +
2959                                                 new->br_blockcount +
2960                                                 right.br_blockcount,
2961                                         left.br_state);
2962                         if (error)
2963                                 goto done;
2964                 }
2965                 break;
2966
2967         case BMAP_LEFT_CONTIG:
2968                 /*
2969                  * New allocation is contiguous with a real allocation
2970                  * on the left.
2971                  * Merge the new allocation with the left neighbor.
2972                  */
2973                 --*idx;
2974                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2975                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2976                         left.br_blockcount + new->br_blockcount);
2977                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2978
2979                 if (cur == NULL) {
2980                         rval = xfs_ilog_fext(whichfork);
2981                 } else {
2982                         rval = 0;
2983                         error = xfs_bmbt_lookup_eq(cur, left.br_startoff,
2984                                         left.br_startblock, left.br_blockcount,
2985                                         &i);
2986                         if (error)
2987                                 goto done;
2988                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2989                         error = xfs_bmbt_update(cur, left.br_startoff,
2990                                         left.br_startblock,
2991                                         left.br_blockcount +
2992                                                 new->br_blockcount,
2993                                         left.br_state);
2994                         if (error)
2995                                 goto done;
2996                 }
2997                 break;
2998
2999         case BMAP_RIGHT_CONTIG:
3000                 /*
3001                  * New allocation is contiguous with a real allocation
3002                  * on the right.
3003                  * Merge the new allocation with the right neighbor.
3004                  */
3005                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
3006                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
3007                         new->br_startoff, new->br_startblock,
3008                         new->br_blockcount + right.br_blockcount,
3009                         right.br_state);
3010                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3011
3012                 if (cur == NULL) {
3013                         rval = xfs_ilog_fext(whichfork);
3014                 } else {
3015                         rval = 0;
3016                         error = xfs_bmbt_lookup_eq(cur,
3017                                         right.br_startoff,
3018                                         right.br_startblock,
3019                                         right.br_blockcount, &i);
3020                         if (error)
3021                                 goto done;
3022                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3023                         error = xfs_bmbt_update(cur, new->br_startoff,
3024                                         new->br_startblock,
3025                                         new->br_blockcount +
3026                                                 right.br_blockcount,
3027                                         right.br_state);
3028                         if (error)
3029                                 goto done;
3030                 }
3031                 break;
3032
3033         case 0:
3034                 /*
3035                  * New allocation is not contiguous with another
3036                  * real allocation.
3037                  * Insert a new entry.
3038                  */
3039                 xfs_iext_insert(ip, *idx, 1, new, state);
3040                 XFS_IFORK_NEXT_SET(ip, whichfork,
3041                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3042                 if (cur == NULL) {
3043                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3044                 } else {
3045                         rval = XFS_ILOG_CORE;
3046                         error = xfs_bmbt_lookup_eq(cur,
3047                                         new->br_startoff,
3048                                         new->br_startblock,
3049                                         new->br_blockcount, &i);
3050                         if (error)
3051                                 goto done;
3052                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
3053                         cur->bc_rec.b.br_state = new->br_state;
3054                         error = xfs_btree_insert(cur, &i);
3055                         if (error)
3056                                 goto done;
3057                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3058                 }
3059                 break;
3060         }
3061
3062         /* add reverse mapping */
3063         error = xfs_rmap_map_extent(mp, dfops, ip, whichfork, new);
3064         if (error)
3065                 goto done;
3066
3067         /* convert to a btree if necessary */
3068         if (xfs_bmap_needs_btree(ip, whichfork)) {
3069                 int     tmp_logflags;   /* partial log flag return val */
3070
3071                 ASSERT(cur == NULL);
3072                 error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, curp,
3073                                 0, &tmp_logflags, whichfork);
3074                 *logflagsp |= tmp_logflags;
3075                 cur = *curp;
3076                 if (error)
3077                         goto done;
3078         }
3079
3080         /* clear out the allocated field, done with it now in any case. */
3081         if (cur)
3082                 cur->bc_private.b.allocated = 0;
3083
3084         xfs_bmap_check_leaf_extents(cur, ip, whichfork);
3085 done:
3086         *logflagsp |= rval;
3087         return error;
3088 }
3089
3090 /*
3091  * Functions used in the extent read, allocate and remove paths
3092  */
3093
3094 /*
3095  * Adjust the size of the new extent based on di_extsize and rt extsize.
3096  */
3097 int
3098 xfs_bmap_extsize_align(
3099         xfs_mount_t     *mp,
3100         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
3101         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
3102         xfs_extlen_t    extsz,          /* align to this extent size */
3103         int             rt,             /* is this a realtime inode? */
3104         int             eof,            /* is extent at end-of-file? */
3105         int             delay,          /* creating delalloc extent? */
3106         int             convert,        /* overwriting unwritten extent? */
3107         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
3108         xfs_extlen_t    *lenp)          /* in/out: aligned length */
3109 {
3110         xfs_fileoff_t   orig_off;       /* original offset */
3111         xfs_extlen_t    orig_alen;      /* original length */
3112         xfs_fileoff_t   orig_end;       /* original off+len */
3113         xfs_fileoff_t   nexto;          /* next file offset */
3114         xfs_fileoff_t   prevo;          /* previous file offset */
3115         xfs_fileoff_t   align_off;      /* temp for offset */
3116         xfs_extlen_t    align_alen;     /* temp for length */
3117         xfs_extlen_t    temp;           /* temp for calculations */
3118
3119         if (convert)
3120                 return 0;
3121
3122         orig_off = align_off = *offp;
3123         orig_alen = align_alen = *lenp;
3124         orig_end = orig_off + orig_alen;
3125
3126         /*
3127          * If this request overlaps an existing extent, then don't
3128          * attempt to perform any additional alignment.
3129          */
3130         if (!delay && !eof &&
3131             (orig_off >= gotp->br_startoff) &&
3132             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3133                 return 0;
3134         }
3135
3136         /*
3137          * If the file offset is unaligned vs. the extent size
3138          * we need to align it.  This will be possible unless
3139          * the file was previously written with a kernel that didn't
3140          * perform this alignment, or if a truncate shot us in the
3141          * foot.
3142          */
3143         temp = do_mod(orig_off, extsz);
3144         if (temp) {
3145                 align_alen += temp;
3146                 align_off -= temp;
3147         }
3148
3149         /* Same adjustment for the end of the requested area. */
3150         temp = (align_alen % extsz);
3151         if (temp)
3152                 align_alen += extsz - temp;
3153
3154         /*
3155          * For large extent hint sizes, the aligned extent might be larger than
3156          * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3157          * the length back under MAXEXTLEN. The outer allocation loops handle
3158          * short allocation just fine, so it is safe to do this. We only want to
3159          * do it when we are forced to, though, because it means more allocation
3160          * operations are required.
3161          */
3162         while (align_alen > MAXEXTLEN)
3163                 align_alen -= extsz;
3164         ASSERT(align_alen <= MAXEXTLEN);
3165
3166         /*
3167          * If the previous block overlaps with this proposed allocation
3168          * then move the start forward without adjusting the length.
3169          */
3170         if (prevp->br_startoff != NULLFILEOFF) {
3171                 if (prevp->br_startblock == HOLESTARTBLOCK)
3172                         prevo = prevp->br_startoff;
3173                 else
3174                         prevo = prevp->br_startoff + prevp->br_blockcount;
3175         } else
3176                 prevo = 0;
3177         if (align_off != orig_off && align_off < prevo)
3178                 align_off = prevo;
3179         /*
3180          * If the next block overlaps with this proposed allocation
3181          * then move the start back without adjusting the length,
3182          * but not before offset 0.
3183          * This may of course make the start overlap previous block,
3184          * and if we hit the offset 0 limit then the next block
3185          * can still overlap too.
3186          */
3187         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3188                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3189                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3190                         nexto = gotp->br_startoff + gotp->br_blockcount;
3191                 else
3192                         nexto = gotp->br_startoff;
3193         } else
3194                 nexto = NULLFILEOFF;
3195         if (!eof &&
3196             align_off + align_alen != orig_end &&
3197             align_off + align_alen > nexto)
3198                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3199         /*
3200          * If we're now overlapping the next or previous extent that
3201          * means we can't fit an extsz piece in this hole.  Just move
3202          * the start forward to the first valid spot and set
3203          * the length so we hit the end.
3204          */
3205         if (align_off != orig_off && align_off < prevo)
3206                 align_off = prevo;
3207         if (align_off + align_alen != orig_end &&
3208             align_off + align_alen > nexto &&
3209             nexto != NULLFILEOFF) {
3210                 ASSERT(nexto > prevo);
3211                 align_alen = nexto - align_off;
3212         }
3213
3214         /*
3215          * If realtime, and the result isn't a multiple of the realtime
3216          * extent size we need to remove blocks until it is.
3217          */
3218         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3219                 /*
3220                  * We're not covering the original request, or
3221                  * we won't be able to once we fix the length.
3222                  */
3223                 if (orig_off < align_off ||
3224                     orig_end > align_off + align_alen ||
3225                     align_alen - temp < orig_alen)
3226                         return -EINVAL;
3227                 /*
3228                  * Try to fix it by moving the start up.
3229                  */
3230                 if (align_off + temp <= orig_off) {
3231                         align_alen -= temp;
3232                         align_off += temp;
3233                 }
3234                 /*
3235                  * Try to fix it by moving the end in.
3236                  */
3237                 else if (align_off + align_alen - temp >= orig_end)
3238                         align_alen -= temp;
3239                 /*
3240                  * Set the start to the minimum then trim the length.
3241                  */
3242                 else {
3243                         align_alen -= orig_off - align_off;
3244                         align_off = orig_off;
3245                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3246                 }
3247                 /*
3248                  * Result doesn't cover the request, fail it.
3249                  */
3250                 if (orig_off < align_off || orig_end > align_off + align_alen)
3251                         return -EINVAL;
3252         } else {
3253                 ASSERT(orig_off >= align_off);
3254                 /* see MAXEXTLEN handling above */
3255                 ASSERT(orig_end <= align_off + align_alen ||
3256                        align_alen + extsz > MAXEXTLEN);
3257         }
3258
3259 #ifdef DEBUG
3260         if (!eof && gotp->br_startoff != NULLFILEOFF)
3261                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3262         if (prevp->br_startoff != NULLFILEOFF)
3263                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3264 #endif
3265
3266         *lenp = align_alen;
3267         *offp = align_off;
3268         return 0;
3269 }
3270
3271 #define XFS_ALLOC_GAP_UNITS     4
3272
3273 void
3274 xfs_bmap_adjacent(
3275         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3276 {
3277         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3278         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3279         xfs_mount_t     *mp;            /* mount point structure */
3280         int             nullfb;         /* true if ap->firstblock isn't set */
3281         int             rt;             /* true if inode is realtime */
3282
3283 #define ISVALID(x,y)    \
3284         (rt ? \
3285                 (x) < mp->m_sb.sb_rblocks : \
3286                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3287                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3288                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3289
3290         mp = ap->ip->i_mount;
3291         nullfb = *ap->firstblock == NULLFSBLOCK;
3292         rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3293                 xfs_alloc_is_userdata(ap->datatype);
3294         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3295         /*
3296          * If allocating at eof, and there's a previous real block,
3297          * try to use its last block as our starting point.
3298          */
3299         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3300             !isnullstartblock(ap->prev.br_startblock) &&
3301             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3302                     ap->prev.br_startblock)) {
3303                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3304                 /*
3305                  * Adjust for the gap between prevp and us.
3306                  */
3307                 adjust = ap->offset -
3308                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3309                 if (adjust &&
3310                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3311                         ap->blkno += adjust;
3312         }
3313         /*
3314          * If not at eof, then compare the two neighbor blocks.
3315          * Figure out whether either one gives us a good starting point,
3316          * and pick the better one.
3317          */
3318         else if (!ap->eof) {
3319                 xfs_fsblock_t   gotbno;         /* right side block number */
3320                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3321                 xfs_fsblock_t   prevbno;        /* left side block number */
3322                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3323
3324                 /*
3325                  * If there's a previous (left) block, select a requested
3326                  * start block based on it.
3327                  */
3328                 if (ap->prev.br_startoff != NULLFILEOFF &&
3329                     !isnullstartblock(ap->prev.br_startblock) &&
3330                     (prevbno = ap->prev.br_startblock +
3331                                ap->prev.br_blockcount) &&
3332                     ISVALID(prevbno, ap->prev.br_startblock)) {
3333                         /*
3334                          * Calculate gap to end of previous block.
3335                          */
3336                         adjust = prevdiff = ap->offset -
3337                                 (ap->prev.br_startoff +
3338                                  ap->prev.br_blockcount);
3339                         /*
3340                          * Figure the startblock based on the previous block's
3341                          * end and the gap size.
3342                          * Heuristic!
3343                          * If the gap is large relative to the piece we're
3344                          * allocating, or using it gives us an invalid block
3345                          * number, then just use the end of the previous block.
3346                          */
3347                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3348                             ISVALID(prevbno + prevdiff,
3349                                     ap->prev.br_startblock))
3350                                 prevbno += adjust;
3351                         else
3352                                 prevdiff += adjust;
3353                         /*
3354                          * If the firstblock forbids it, can't use it,
3355                          * must use default.
3356                          */
3357                         if (!rt && !nullfb &&
3358                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3359                                 prevbno = NULLFSBLOCK;
3360                 }
3361                 /*
3362                  * No previous block or can't follow it, just default.
3363                  */
3364                 else
3365                         prevbno = NULLFSBLOCK;
3366                 /*
3367                  * If there's a following (right) block, select a requested
3368                  * start block based on it.
3369                  */
3370                 if (!isnullstartblock(ap->got.br_startblock)) {
3371                         /*
3372                          * Calculate gap to start of next block.
3373                          */
3374                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3375                         /*
3376                          * Figure the startblock based on the next block's
3377                          * start and the gap size.
3378                          */
3379                         gotbno = ap->got.br_startblock;
3380                         /*
3381                          * Heuristic!
3382                          * If the gap is large relative to the piece we're
3383                          * allocating, or using it gives us an invalid block
3384                          * number, then just use the start of the next block
3385                          * offset by our length.
3386                          */
3387                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3388                             ISVALID(gotbno - gotdiff, gotbno))
3389                                 gotbno -= adjust;
3390                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3391                                 gotbno -= ap->length;
3392                                 gotdiff += adjust - ap->length;
3393                         } else
3394                                 gotdiff += adjust;
3395                         /*
3396                          * If the firstblock forbids it, can't use it,
3397                          * must use default.
3398                          */
3399                         if (!rt && !nullfb &&
3400                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3401                                 gotbno = NULLFSBLOCK;
3402                 }
3403                 /*
3404                  * No next block, just default.
3405                  */
3406                 else
3407                         gotbno = NULLFSBLOCK;
3408                 /*
3409                  * If both valid, pick the better one, else the only good
3410                  * one, else ap->blkno is already set (to 0 or the inode block).
3411                  */
3412                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3413                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3414                 else if (prevbno != NULLFSBLOCK)
3415                         ap->blkno = prevbno;
3416                 else if (gotbno != NULLFSBLOCK)
3417                         ap->blkno = gotbno;
3418         }
3419 #undef ISVALID
3420 }
3421
3422 static int
3423 xfs_bmap_longest_free_extent(
3424         struct xfs_trans        *tp,
3425         xfs_agnumber_t          ag,
3426         xfs_extlen_t            *blen,
3427         int                     *notinit)
3428 {
3429         struct xfs_mount        *mp = tp->t_mountp;
3430         struct xfs_perag        *pag;
3431         xfs_extlen_t            longest;
3432         int                     error = 0;
3433
3434         pag = xfs_perag_get(mp, ag);
3435         if (!pag->pagf_init) {
3436                 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3437                 if (error)
3438                         goto out;
3439
3440                 if (!pag->pagf_init) {
3441                         *notinit = 1;
3442                         goto out;
3443                 }
3444         }
3445
3446         longest = xfs_alloc_longest_free_extent(mp, pag,
3447                                 xfs_alloc_min_freelist(mp, pag),
3448                                 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3449         if (*blen < longest)
3450                 *blen = longest;
3451
3452 out:
3453         xfs_perag_put(pag);
3454         return error;
3455 }
3456
3457 static void
3458 xfs_bmap_select_minlen(
3459         struct xfs_bmalloca     *ap,
3460         struct xfs_alloc_arg    *args,
3461         xfs_extlen_t            *blen,
3462         int                     notinit)
3463 {
3464         if (notinit || *blen < ap->minlen) {
3465                 /*
3466                  * Since we did a BUF_TRYLOCK above, it is possible that
3467                  * there is space for this request.
3468                  */
3469                 args->minlen = ap->minlen;
3470         } else if (*blen < args->maxlen) {
3471                 /*
3472                  * If the best seen length is less than the request length,
3473                  * use the best as the minimum.
3474                  */
3475                 args->minlen = *blen;
3476         } else {
3477                 /*
3478                  * Otherwise we've seen an extent as big as maxlen, use that
3479                  * as the minimum.
3480                  */
3481                 args->minlen = args->maxlen;
3482         }
3483 }
3484
3485 STATIC int
3486 xfs_bmap_btalloc_nullfb(
3487         struct xfs_bmalloca     *ap,
3488         struct xfs_alloc_arg    *args,
3489         xfs_extlen_t            *blen)
3490 {
3491         struct xfs_mount        *mp = ap->ip->i_mount;
3492         xfs_agnumber_t          ag, startag;
3493         int                     notinit = 0;
3494         int                     error;
3495
3496         args->type = XFS_ALLOCTYPE_START_BNO;
3497         args->total = ap->total;
3498
3499         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3500         if (startag == NULLAGNUMBER)
3501                 startag = ag = 0;
3502
3503         while (*blen < args->maxlen) {
3504                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3505                                                      &notinit);
3506                 if (error)
3507                         return error;
3508
3509                 if (++ag == mp->m_sb.sb_agcount)
3510                         ag = 0;
3511                 if (ag == startag)
3512                         break;
3513         }
3514
3515         xfs_bmap_select_minlen(ap, args, blen, notinit);
3516         return 0;
3517 }
3518
3519 STATIC int
3520 xfs_bmap_btalloc_filestreams(
3521         struct xfs_bmalloca     *ap,
3522         struct xfs_alloc_arg    *args,
3523         xfs_extlen_t            *blen)
3524 {
3525         struct xfs_mount        *mp = ap->ip->i_mount;
3526         xfs_agnumber_t          ag;
3527         int                     notinit = 0;
3528         int                     error;
3529
3530         args->type = XFS_ALLOCTYPE_NEAR_BNO;
3531         args->total = ap->total;
3532
3533         ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3534         if (ag == NULLAGNUMBER)
3535                 ag = 0;
3536
3537         error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3538         if (error)
3539                 return error;
3540
3541         if (*blen < args->maxlen) {
3542                 error = xfs_filestream_new_ag(ap, &ag);
3543                 if (error)
3544                         return error;
3545
3546                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3547                                                      &notinit);
3548                 if (error)
3549                         return error;
3550
3551         }
3552
3553         xfs_bmap_select_minlen(ap, args, blen, notinit);
3554
3555         /*
3556          * Set the failure fallback case to look in the selected AG as stream
3557          * may have moved.
3558          */
3559         ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3560         return 0;
3561 }
3562
3563 STATIC int
3564 xfs_bmap_btalloc(
3565         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3566 {
3567         xfs_mount_t     *mp;            /* mount point structure */
3568         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3569         xfs_extlen_t    align = 0;      /* minimum allocation alignment */
3570         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3571         xfs_agnumber_t  ag;
3572         xfs_alloc_arg_t args;
3573         xfs_extlen_t    blen;
3574         xfs_extlen_t    nextminlen = 0;
3575         int             nullfb;         /* true if ap->firstblock isn't set */
3576         int             isaligned;
3577         int             tryagain;
3578         int             error;
3579         int             stripe_align;
3580
3581         ASSERT(ap->length);
3582
3583         mp = ap->ip->i_mount;
3584
3585         /* stripe alignment for allocation is determined by mount parameters */
3586         stripe_align = 0;
3587         if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3588                 stripe_align = mp->m_swidth;
3589         else if (mp->m_dalign)
3590                 stripe_align = mp->m_dalign;
3591
3592         if (ap->flags & XFS_BMAPI_COWFORK)
3593                 align = xfs_get_cowextsz_hint(ap->ip);
3594         else if (xfs_alloc_is_userdata(ap->datatype))
3595                 align = xfs_get_extsz_hint(ap->ip);
3596         if (align) {
3597                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3598                                                 align, 0, ap->eof, 0, ap->conv,
3599                                                 &ap->offset, &ap->length);
3600                 ASSERT(!error);
3601                 ASSERT(ap->length);
3602         }
3603
3604
3605         nullfb = *ap->firstblock == NULLFSBLOCK;
3606         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3607         if (nullfb) {
3608                 if (xfs_alloc_is_userdata(ap->datatype) &&
3609                     xfs_inode_is_filestream(ap->ip)) {
3610                         ag = xfs_filestream_lookup_ag(ap->ip);
3611                         ag = (ag != NULLAGNUMBER) ? ag : 0;
3612                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3613                 } else {
3614                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3615                 }
3616         } else
3617                 ap->blkno = *ap->firstblock;
3618
3619         xfs_bmap_adjacent(ap);
3620
3621         /*
3622          * If allowed, use ap->blkno; otherwise must use firstblock since
3623          * it's in the right allocation group.
3624          */
3625         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3626                 ;
3627         else
3628                 ap->blkno = *ap->firstblock;
3629         /*
3630          * Normal allocation, done through xfs_alloc_vextent.
3631          */
3632         tryagain = isaligned = 0;
3633         memset(&args, 0, sizeof(args));
3634         args.tp = ap->tp;
3635         args.mp = mp;
3636         args.fsbno = ap->blkno;
3637         xfs_rmap_skip_owner_update(&args.oinfo);
3638
3639         /* Trim the allocation back to the maximum an AG can fit. */
3640         args.maxlen = MIN(ap->length, mp->m_ag_max_usable);
3641         args.firstblock = *ap->firstblock;
3642         blen = 0;
3643         if (nullfb) {
3644                 /*
3645                  * Search for an allocation group with a single extent large
3646                  * enough for the request.  If one isn't found, then adjust
3647                  * the minimum allocation size to the largest space found.
3648                  */
3649                 if (xfs_alloc_is_userdata(ap->datatype) &&
3650                     xfs_inode_is_filestream(ap->ip))
3651                         error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3652                 else
3653                         error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3654                 if (error)
3655                         return error;
3656         } else if (ap->dfops->dop_low) {
3657                 if (xfs_inode_is_filestream(ap->ip))
3658                         args.type = XFS_ALLOCTYPE_FIRST_AG;
3659                 else
3660                         args.type = XFS_ALLOCTYPE_START_BNO;
3661                 args.total = args.minlen = ap->minlen;
3662         } else {
3663                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3664                 args.total = ap->total;
3665                 args.minlen = ap->minlen;
3666         }
3667         /* apply extent size hints if obtained earlier */
3668         if (align) {
3669                 args.prod = align;
3670                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3671                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3672         } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3673                 args.prod = 1;
3674                 args.mod = 0;
3675         } else {
3676                 args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3677                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3678                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3679         }
3680         /*
3681          * If we are not low on available data blocks, and the
3682          * underlying logical volume manager is a stripe, and
3683          * the file offset is zero then try to allocate data
3684          * blocks on stripe unit boundary.
3685          * NOTE: ap->aeof is only set if the allocation length
3686          * is >= the stripe unit and the allocation offset is
3687          * at the end of file.
3688          */
3689         if (!ap->dfops->dop_low && ap->aeof) {
3690                 if (!ap->offset) {
3691                         args.alignment = stripe_align;
3692                         atype = args.type;
3693                         isaligned = 1;
3694                         /*
3695                          * Adjust for alignment
3696                          */
3697                         if (blen > args.alignment && blen <= args.maxlen)
3698                                 args.minlen = blen - args.alignment;
3699                         args.minalignslop = 0;
3700                 } else {
3701                         /*
3702                          * First try an exact bno allocation.
3703                          * If it fails then do a near or start bno
3704                          * allocation with alignment turned on.
3705                          */
3706                         atype = args.type;
3707                         tryagain = 1;
3708                         args.type = XFS_ALLOCTYPE_THIS_BNO;
3709                         args.alignment = 1;
3710                         /*
3711                          * Compute the minlen+alignment for the
3712                          * next case.  Set slop so that the value
3713                          * of minlen+alignment+slop doesn't go up
3714                          * between the calls.
3715                          */
3716                         if (blen > stripe_align && blen <= args.maxlen)
3717                                 nextminlen = blen - stripe_align;
3718                         else
3719                                 nextminlen = args.minlen;
3720                         if (nextminlen + stripe_align > args.minlen + 1)
3721                                 args.minalignslop =
3722                                         nextminlen + stripe_align -
3723                                         args.minlen - 1;
3724                         else
3725                                 args.minalignslop = 0;
3726                 }
3727         } else {
3728                 args.alignment = 1;
3729                 args.minalignslop = 0;
3730         }
3731         args.minleft = ap->minleft;
3732         args.wasdel = ap->wasdel;
3733         args.resv = XFS_AG_RESV_NONE;
3734         args.datatype = ap->datatype;
3735         if (ap->datatype & XFS_ALLOC_USERDATA_ZERO)
3736                 args.ip = ap->ip;
3737
3738         error = xfs_alloc_vextent(&args);
3739         if (error)
3740                 return error;
3741
3742         if (tryagain && args.fsbno == NULLFSBLOCK) {
3743                 /*
3744                  * Exact allocation failed. Now try with alignment
3745                  * turned on.
3746                  */
3747                 args.type = atype;
3748                 args.fsbno = ap->blkno;
3749                 args.alignment = stripe_align;
3750                 args.minlen = nextminlen;
3751                 args.minalignslop = 0;
3752                 isaligned = 1;
3753                 if ((error = xfs_alloc_vextent(&args)))
3754                         return error;
3755         }
3756         if (isaligned && args.fsbno == NULLFSBLOCK) {
3757                 /*
3758                  * allocation failed, so turn off alignment and
3759                  * try again.
3760                  */
3761                 args.type = atype;
3762                 args.fsbno = ap->blkno;
3763                 args.alignment = 0;
3764                 if ((error = xfs_alloc_vextent(&args)))
3765                         return error;
3766         }
3767         if (args.fsbno == NULLFSBLOCK && nullfb &&
3768             args.minlen > ap->minlen) {
3769                 args.minlen = ap->minlen;
3770                 args.type = XFS_ALLOCTYPE_START_BNO;
3771                 args.fsbno = ap->blkno;
3772                 if ((error = xfs_alloc_vextent(&args)))
3773                         return error;
3774         }
3775         if (args.fsbno == NULLFSBLOCK && nullfb) {
3776                 args.fsbno = 0;
3777                 args.type = XFS_ALLOCTYPE_FIRST_AG;
3778                 args.total = ap->minlen;
3779                 if ((error = xfs_alloc_vextent(&args)))
3780                         return error;
3781                 ap->dfops->dop_low = true;
3782         }
3783         if (args.fsbno != NULLFSBLOCK) {
3784                 /*
3785                  * check the allocation happened at the same or higher AG than
3786                  * the first block that was allocated.
3787                  */
3788                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3789                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) <=
3790                        XFS_FSB_TO_AGNO(mp, args.fsbno));
3791
3792                 ap->blkno = args.fsbno;
3793                 if (*ap->firstblock == NULLFSBLOCK)
3794                         *ap->firstblock = args.fsbno;
3795                 ASSERT(nullfb || fb_agno <= args.agno);
3796                 ap->length = args.len;
3797                 if (!(ap->flags & XFS_BMAPI_COWFORK))
3798                         ap->ip->i_d.di_nblocks += args.len;
3799                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3800                 if (ap->wasdel)
3801                         ap->ip->i_delayed_blks -= args.len;
3802                 /*
3803                  * Adjust the disk quota also. This was reserved
3804                  * earlier.
3805                  */
3806                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3807                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3808                                         XFS_TRANS_DQ_BCOUNT,
3809                         (long) args.len);
3810         } else {
3811                 ap->blkno = NULLFSBLOCK;
3812                 ap->length = 0;
3813         }
3814         return 0;
3815 }
3816
3817 /*
3818  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3819  * It figures out where to ask the underlying allocator to put the new extent.
3820  */
3821 STATIC int
3822 xfs_bmap_alloc(
3823         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3824 {
3825         if (XFS_IS_REALTIME_INODE(ap->ip) &&
3826             xfs_alloc_is_userdata(ap->datatype))
3827                 return xfs_bmap_rtalloc(ap);
3828         return xfs_bmap_btalloc(ap);
3829 }
3830
3831 /* Trim extent to fit a logical block range. */
3832 void
3833 xfs_trim_extent(
3834         struct xfs_bmbt_irec    *irec,
3835         xfs_fileoff_t           bno,
3836         xfs_filblks_t           len)
3837 {
3838         xfs_fileoff_t           distance;
3839         xfs_fileoff_t           end = bno + len;
3840
3841         if (irec->br_startoff + irec->br_blockcount <= bno ||
3842             irec->br_startoff >= end) {
3843                 irec->br_blockcount = 0;
3844                 return;
3845         }
3846
3847         if (irec->br_startoff < bno) {
3848                 distance = bno - irec->br_startoff;
3849                 if (isnullstartblock(irec->br_startblock))
3850                         irec->br_startblock = DELAYSTARTBLOCK;
3851                 if (irec->br_startblock != DELAYSTARTBLOCK &&
3852                     irec->br_startblock != HOLESTARTBLOCK)
3853                         irec->br_startblock += distance;
3854                 irec->br_startoff += distance;
3855                 irec->br_blockcount -= distance;
3856         }
3857
3858         if (end < irec->br_startoff + irec->br_blockcount) {
3859                 distance = irec->br_startoff + irec->br_blockcount - end;
3860                 irec->br_blockcount -= distance;
3861         }
3862 }
3863
3864 /*
3865  * Trim the returned map to the required bounds
3866  */
3867 STATIC void
3868 xfs_bmapi_trim_map(
3869         struct xfs_bmbt_irec    *mval,
3870         struct xfs_bmbt_irec    *got,
3871         xfs_fileoff_t           *bno,
3872         xfs_filblks_t           len,
3873         xfs_fileoff_t           obno,
3874         xfs_fileoff_t           end,
3875         int                     n,
3876         int                     flags)
3877 {
3878         if ((flags & XFS_BMAPI_ENTIRE) ||
3879             got->br_startoff + got->br_blockcount <= obno) {
3880                 *mval = *got;
3881                 if (isnullstartblock(got->br_startblock))
3882                         mval->br_startblock = DELAYSTARTBLOCK;
3883                 return;
3884         }
3885
3886         if (obno > *bno)
3887                 *bno = obno;
3888         ASSERT((*bno >= obno) || (n == 0));
3889         ASSERT(*bno < end);
3890         mval->br_startoff = *bno;
3891         if (isnullstartblock(got->br_startblock))
3892                 mval->br_startblock = DELAYSTARTBLOCK;
3893         else
3894                 mval->br_startblock = got->br_startblock +
3895                                         (*bno - got->br_startoff);
3896         /*
3897          * Return the minimum of what we got and what we asked for for
3898          * the length.  We can use the len variable here because it is
3899          * modified below and we could have been there before coming
3900          * here if the first part of the allocation didn't overlap what
3901          * was asked for.
3902          */
3903         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3904                         got->br_blockcount - (*bno - got->br_startoff));
3905         mval->br_state = got->br_state;
3906         ASSERT(mval->br_blockcount <= len);
3907         return;
3908 }
3909
3910 /*
3911  * Update and validate the extent map to return
3912  */
3913 STATIC void
3914 xfs_bmapi_update_map(
3915         struct xfs_bmbt_irec    **map,
3916         xfs_fileoff_t           *bno,
3917         xfs_filblks_t           *len,
3918         xfs_fileoff_t           obno,
3919         xfs_fileoff_t           end,
3920         int                     *n,
3921         int                     flags)
3922 {
3923         xfs_bmbt_irec_t *mval = *map;
3924
3925         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3926                ((mval->br_startoff + mval->br_blockcount) <= end));
3927         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3928                (mval->br_startoff < obno));
3929
3930         *bno = mval->br_startoff + mval->br_blockcount;
3931         *len = end - *bno;
3932         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3933                 /* update previous map with new information */
3934                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3935                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3936                 ASSERT(mval->br_state == mval[-1].br_state);
3937                 mval[-1].br_blockcount = mval->br_blockcount;
3938                 mval[-1].br_state = mval->br_state;
3939         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3940                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
3941                    mval[-1].br_startblock != HOLESTARTBLOCK &&
3942                    mval->br_startblock == mval[-1].br_startblock +
3943                                           mval[-1].br_blockcount &&
3944                    ((flags & XFS_BMAPI_IGSTATE) ||
3945                         mval[-1].br_state == mval->br_state)) {
3946                 ASSERT(mval->br_startoff ==
3947                        mval[-1].br_startoff + mval[-1].br_blockcount);
3948                 mval[-1].br_blockcount += mval->br_blockcount;
3949         } else if (*n > 0 &&
3950                    mval->br_startblock == DELAYSTARTBLOCK &&
3951                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
3952                    mval->br_startoff ==
3953                    mval[-1].br_startoff + mval[-1].br_blockcount) {
3954                 mval[-1].br_blockcount += mval->br_blockcount;
3955                 mval[-1].br_state = mval->br_state;
3956         } else if (!((*n == 0) &&
3957                      ((mval->br_startoff + mval->br_blockcount) <=
3958                       obno))) {
3959                 mval++;
3960                 (*n)++;
3961         }
3962         *map = mval;
3963 }
3964
3965 /*
3966  * Map file blocks to filesystem blocks without allocation.
3967  */
3968 int
3969 xfs_bmapi_read(
3970         struct xfs_inode        *ip,
3971         xfs_fileoff_t           bno,
3972         xfs_filblks_t           len,
3973         struct xfs_bmbt_irec    *mval,
3974         int                     *nmap,
3975         int                     flags)
3976 {
3977         struct xfs_mount        *mp = ip->i_mount;
3978         struct xfs_ifork        *ifp;
3979         struct xfs_bmbt_irec    got;
3980         xfs_fileoff_t           obno;
3981         xfs_fileoff_t           end;
3982         xfs_extnum_t            idx;
3983         int                     error;
3984         bool                    eof = false;
3985         int                     n = 0;
3986         int                     whichfork = xfs_bmapi_whichfork(flags);
3987
3988         ASSERT(*nmap >= 1);
3989         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
3990                            XFS_BMAPI_IGSTATE|XFS_BMAPI_COWFORK)));
3991         ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3992
3993         if (unlikely(XFS_TEST_ERROR(
3994             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3995              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
3996              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
3997                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
3998                 return -EFSCORRUPTED;
3999         }
4000
4001         if (XFS_FORCED_SHUTDOWN(mp))
4002                 return -EIO;
4003
4004         XFS_STATS_INC(mp, xs_blk_mapr);
4005
4006         ifp = XFS_IFORK_PTR(ip, whichfork);
4007
4008         /* No CoW fork?  Return a hole. */
4009         if (whichfork == XFS_COW_FORK && !ifp) {
4010                 mval->br_startoff = bno;
4011                 mval->br_startblock = HOLESTARTBLOCK;
4012                 mval->br_blockcount = len;
4013                 mval->br_state = XFS_EXT_NORM;
4014                 *nmap = 1;
4015                 return 0;
4016         }
4017
4018         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4019                 error = xfs_iread_extents(NULL, ip, whichfork);
4020                 if (error)
4021                         return error;
4022         }
4023
4024         if (!xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got))
4025                 eof = true;
4026         end = bno + len;
4027         obno = bno;
4028
4029         while (bno < end && n < *nmap) {
4030                 /* Reading past eof, act as though there's a hole up to end. */
4031                 if (eof)
4032                         got.br_startoff = end;
4033                 if (got.br_startoff > bno) {
4034                         /* Reading in a hole.  */
4035                         mval->br_startoff = bno;
4036                         mval->br_startblock = HOLESTARTBLOCK;
4037                         mval->br_blockcount =
4038                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4039                         mval->br_state = XFS_EXT_NORM;
4040                         bno += mval->br_blockcount;
4041                         len -= mval->br_blockcount;
4042                         mval++;
4043                         n++;
4044                         continue;
4045                 }
4046
4047                 /* set up the extent map to return. */
4048                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4049                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4050
4051                 /* If we're done, stop now. */
4052                 if (bno >= end || n >= *nmap)
4053                         break;
4054
4055                 /* Else go on to the next record. */
4056                 if (!xfs_iext_get_extent(ifp, ++idx, &got))
4057                         eof = true;
4058         }
4059         *nmap = n;
4060         return 0;
4061 }
4062
4063 /*
4064  * Add a delayed allocation extent to an inode. Blocks are reserved from the
4065  * global pool and the extent inserted into the inode in-core extent tree.
4066  *
4067  * On entry, got refers to the first extent beyond the offset of the extent to
4068  * allocate or eof is specified if no such extent exists. On return, got refers
4069  * to the extent record that was inserted to the inode fork.
4070  *
4071  * Note that the allocated extent may have been merged with contiguous extents
4072  * during insertion into the inode fork. Thus, got does not reflect the current
4073  * state of the inode fork on return. If necessary, the caller can use lastx to
4074  * look up the updated record in the inode fork.
4075  */
4076 int
4077 xfs_bmapi_reserve_delalloc(
4078         struct xfs_inode        *ip,
4079         int                     whichfork,
4080         xfs_fileoff_t           off,
4081         xfs_filblks_t           len,
4082         xfs_filblks_t           prealloc,
4083         struct xfs_bmbt_irec    *got,
4084         xfs_extnum_t            *lastx,
4085         int                     eof)
4086 {
4087         struct xfs_mount        *mp = ip->i_mount;
4088         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4089         xfs_extlen_t            alen;
4090         xfs_extlen_t            indlen;
4091         char                    rt = XFS_IS_REALTIME_INODE(ip);
4092         xfs_extlen_t            extsz;
4093         int                     error;
4094         xfs_fileoff_t           aoff = off;
4095
4096         /*
4097          * Cap the alloc length. Keep track of prealloc so we know whether to
4098          * tag the inode before we return.
4099          */
4100         alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
4101         if (!eof)
4102                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4103         if (prealloc && alen >= len)
4104                 prealloc = alen - len;
4105
4106         /* Figure out the extent size, adjust alen */
4107         if (whichfork == XFS_COW_FORK)
4108                 extsz = xfs_get_cowextsz_hint(ip);
4109         else
4110                 extsz = xfs_get_extsz_hint(ip);
4111         if (extsz) {
4112                 struct xfs_bmbt_irec    prev;
4113
4114                 if (!xfs_iext_get_extent(ifp, *lastx - 1, &prev))
4115                         prev.br_startoff = NULLFILEOFF;
4116
4117                 error = xfs_bmap_extsize_align(mp, got, &prev, extsz, rt, eof,
4118                                                1, 0, &aoff, &alen);
4119                 ASSERT(!error);
4120         }
4121
4122         if (rt)
4123                 extsz = alen / mp->m_sb.sb_rextsize;
4124
4125         /*
4126          * Make a transaction-less quota reservation for delayed allocation
4127          * blocks.  This number gets adjusted later.  We return if we haven't
4128          * allocated blocks already inside this loop.
4129          */
4130         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4131                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4132         if (error)
4133                 return error;
4134
4135         /*
4136          * Split changing sb for alen and indlen since they could be coming
4137          * from different places.
4138          */
4139         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4140         ASSERT(indlen > 0);
4141
4142         if (rt) {
4143                 error = xfs_mod_frextents(mp, -((int64_t)extsz));
4144         } else {
4145                 error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4146         }
4147
4148         if (error)
4149                 goto out_unreserve_quota;
4150
4151         error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4152         if (error)
4153                 goto out_unreserve_blocks;
4154
4155
4156         ip->i_delayed_blks += alen;
4157
4158         got->br_startoff = aoff;
4159         got->br_startblock = nullstartblock(indlen);
4160         got->br_blockcount = alen;
4161         got->br_state = XFS_EXT_NORM;
4162
4163         xfs_bmap_add_extent_hole_delay(ip, whichfork, lastx, got);
4164
4165         /*
4166          * Tag the inode if blocks were preallocated. Note that COW fork
4167          * preallocation can occur at the start or end of the extent, even when
4168          * prealloc == 0, so we must also check the aligned offset and length.
4169          */
4170         if (whichfork == XFS_DATA_FORK && prealloc)
4171                 xfs_inode_set_eofblocks_tag(ip);
4172         if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4173                 xfs_inode_set_cowblocks_tag(ip);
4174
4175         return 0;
4176
4177 out_unreserve_blocks:
4178         if (rt)
4179                 xfs_mod_frextents(mp, extsz);
4180         else
4181                 xfs_mod_fdblocks(mp, alen, false);
4182 out_unreserve_quota:
4183         if (XFS_IS_QUOTA_ON(mp))
4184                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4185                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4186         return error;
4187 }
4188
4189 static int
4190 xfs_bmapi_allocate(
4191         struct xfs_bmalloca     *bma)
4192 {
4193         struct xfs_mount        *mp = bma->ip->i_mount;
4194         int                     whichfork = xfs_bmapi_whichfork(bma->flags);
4195         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4196         int                     tmp_logflags = 0;
4197         int                     error;
4198
4199         ASSERT(bma->length > 0);
4200
4201         /*
4202          * For the wasdelay case, we could also just allocate the stuff asked
4203          * for in this bmap call but that wouldn't be as good.
4204          */
4205         if (bma->wasdel) {
4206                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4207                 bma->offset = bma->got.br_startoff;
4208                 if (bma->idx) {
4209                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4210                                          &bma->prev);
4211                 }
4212         } else {
4213                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4214                 if (!bma->eof)
4215                         bma->length = XFS_FILBLKS_MIN(bma->length,
4216                                         bma->got.br_startoff - bma->offset);
4217         }
4218
4219         /*
4220          * Set the data type being allocated. For the data fork, the first data
4221          * in the file is treated differently to all other allocations. For the
4222          * attribute fork, we only need to ensure the allocated range is not on
4223          * the busy list.
4224          */
4225         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4226                 bma->datatype = XFS_ALLOC_NOBUSY;
4227                 if (whichfork == XFS_DATA_FORK) {
4228                         if (bma->offset == 0)
4229                                 bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4230                         else
4231                                 bma->datatype |= XFS_ALLOC_USERDATA;
4232                 }
4233                 if (bma->flags & XFS_BMAPI_ZERO)
4234                         bma->datatype |= XFS_ALLOC_USERDATA_ZERO;
4235         }
4236
4237         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4238
4239         /*
4240          * Only want to do the alignment at the eof if it is userdata and
4241          * allocation length is larger than a stripe unit.
4242          */
4243         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4244             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4245                 error = xfs_bmap_isaeof(bma, whichfork);
4246                 if (error)
4247                         return error;
4248         }
4249
4250         error = xfs_bmap_alloc(bma);
4251         if (error)
4252                 return error;
4253
4254         if (bma->cur)
4255                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4256         if (bma->blkno == NULLFSBLOCK)
4257                 return 0;
4258         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4259                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4260                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4261                 bma->cur->bc_private.b.dfops = bma->dfops;
4262         }
4263         /*
4264          * Bump the number of extents we've allocated
4265          * in this call.
4266          */
4267         bma->nallocs++;
4268
4269         if (bma->cur)
4270                 bma->cur->bc_private.b.flags =
4271                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4272
4273         bma->got.br_startoff = bma->offset;
4274         bma->got.br_startblock = bma->blkno;
4275         bma->got.br_blockcount = bma->length;
4276         bma->got.br_state = XFS_EXT_NORM;
4277
4278         /*
4279          * In the data fork, a wasdelay extent has been initialized, so
4280          * shouldn't be flagged as unwritten.
4281          *
4282          * For the cow fork, however, we convert delalloc reservations
4283          * (extents allocated for speculative preallocation) to
4284          * allocated unwritten extents, and only convert the unwritten
4285          * extents to real extents when we're about to write the data.
4286          */
4287         if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) &&
4288             (bma->flags & XFS_BMAPI_PREALLOC) &&
4289             xfs_sb_version_hasextflgbit(&mp->m_sb))
4290                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4291
4292         if (bma->wasdel)
4293                 error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4294         else
4295                 error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4296                                 whichfork, &bma->idx, &bma->cur, &bma->got,
4297                                 bma->firstblock, bma->dfops, &bma->logflags);
4298
4299         bma->logflags |= tmp_logflags;
4300         if (error)
4301                 return error;
4302
4303         /*
4304          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4305          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4306          * the neighbouring ones.
4307          */
4308         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4309
4310         ASSERT(bma->got.br_startoff <= bma->offset);
4311         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4312                bma->offset + bma->length);
4313         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4314                bma->got.br_state == XFS_EXT_UNWRITTEN);
4315         return 0;
4316 }
4317
4318 STATIC int
4319 xfs_bmapi_convert_unwritten(
4320         struct xfs_bmalloca     *bma,
4321         struct xfs_bmbt_irec    *mval,
4322         xfs_filblks_t           len,
4323         int                     flags)
4324 {
4325         int                     whichfork = xfs_bmapi_whichfork(flags);
4326         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4327         int                     tmp_logflags = 0;
4328         int                     error;
4329
4330         /* check if we need to do unwritten->real conversion */
4331         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4332             (flags & XFS_BMAPI_PREALLOC))
4333                 return 0;
4334
4335         /* check if we need to do real->unwritten conversion */
4336         if (mval->br_state == XFS_EXT_NORM &&
4337             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4338                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4339                 return 0;
4340
4341         /*
4342          * Modify (by adding) the state flag, if writing.
4343          */
4344         ASSERT(mval->br_blockcount <= len);
4345         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4346                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4347                                         bma->ip, whichfork);
4348                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4349                 bma->cur->bc_private.b.dfops = bma->dfops;
4350         }
4351         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4352                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4353
4354         /*
4355          * Before insertion into the bmbt, zero the range being converted
4356          * if required.
4357          */
4358         if (flags & XFS_BMAPI_ZERO) {
4359                 error = xfs_zero_extent(bma->ip, mval->br_startblock,
4360                                         mval->br_blockcount);
4361                 if (error)
4362                         return error;
4363         }
4364
4365         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4366                         &bma->idx, &bma->cur, mval, bma->firstblock, bma->dfops,
4367                         &tmp_logflags);
4368         /*
4369          * Log the inode core unconditionally in the unwritten extent conversion
4370          * path because the conversion might not have done so (e.g., if the
4371          * extent count hasn't changed). We need to make sure the inode is dirty
4372          * in the transaction for the sake of fsync(), even if nothing has
4373          * changed, because fsync() will not force the log for this transaction
4374          * unless it sees the inode pinned.
4375          *
4376          * Note: If we're only converting cow fork extents, there aren't
4377          * any on-disk updates to make, so we don't need to log anything.
4378          */
4379         if (whichfork != XFS_COW_FORK)
4380                 bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4381         if (error)
4382                 return error;
4383
4384         /*
4385          * Update our extent pointer, given that
4386          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4387          * of the neighbouring ones.
4388          */
4389         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4390
4391         /*
4392          * We may have combined previously unwritten space with written space,
4393          * so generate another request.
4394          */
4395         if (mval->br_blockcount < len)
4396                 return -EAGAIN;
4397         return 0;
4398 }
4399
4400 /*
4401  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4402  * extent state if necessary.  Details behaviour is controlled by the flags
4403  * parameter.  Only allocates blocks from a single allocation group, to avoid
4404  * locking problems.
4405  *
4406  * The returned value in "firstblock" from the first call in a transaction
4407  * must be remembered and presented to subsequent calls in "firstblock".
4408  * An upper bound for the number of blocks to be allocated is supplied to
4409  * the first call in "total"; if no allocation group has that many free
4410  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4411  */
4412 int
4413 xfs_bmapi_write(
4414         struct xfs_trans        *tp,            /* transaction pointer */
4415         struct xfs_inode        *ip,            /* incore inode */
4416         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4417         xfs_filblks_t           len,            /* length to map in file */
4418         int                     flags,          /* XFS_BMAPI_... */
4419         xfs_fsblock_t           *firstblock,    /* first allocated block
4420                                                    controls a.g. for allocs */
4421         xfs_extlen_t            total,          /* total blocks needed */
4422         struct xfs_bmbt_irec    *mval,          /* output: map values */
4423         int                     *nmap,          /* i/o: mval size/count */
4424         struct xfs_defer_ops    *dfops)         /* i/o: list extents to free */
4425 {
4426         struct xfs_mount        *mp = ip->i_mount;
4427         struct xfs_ifork        *ifp;
4428         struct xfs_bmalloca     bma = { NULL }; /* args for xfs_bmap_alloc */
4429         xfs_fileoff_t           end;            /* end of mapped file region */
4430         bool                    eof = false;    /* after the end of extents */
4431         int                     error;          /* error return */
4432         int                     n;              /* current extent index */
4433         xfs_fileoff_t           obno;           /* old block number (offset) */
4434         int                     whichfork;      /* data or attr fork */
4435
4436 #ifdef DEBUG
4437         xfs_fileoff_t           orig_bno;       /* original block number value */
4438         int                     orig_flags;     /* original flags arg value */
4439         xfs_filblks_t           orig_len;       /* original value of len arg */
4440         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4441         int                     orig_nmap;      /* original value of *nmap */
4442
4443         orig_bno = bno;
4444         orig_len = len;
4445         orig_flags = flags;
4446         orig_mval = mval;
4447         orig_nmap = *nmap;
4448 #endif
4449         whichfork = xfs_bmapi_whichfork(flags);
4450
4451         ASSERT(*nmap >= 1);
4452         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4453         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4454         ASSERT(tp != NULL ||
4455                (flags & (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)) ==
4456                         (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK));
4457         ASSERT(len > 0);
4458         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4459         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4460         ASSERT(!(flags & XFS_BMAPI_REMAP));
4461
4462         /* zeroing is for currently only for data extents, not metadata */
4463         ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4464                         (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4465         /*
4466          * we can allocate unwritten extents or pre-zero allocated blocks,
4467          * but it makes no sense to do both at once. This would result in
4468          * zeroing the unwritten extent twice, but it still being an
4469          * unwritten extent....
4470          */
4471         ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4472                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4473
4474         if (unlikely(XFS_TEST_ERROR(
4475             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4476              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4477              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4478                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4479                 return -EFSCORRUPTED;
4480         }
4481
4482         if (XFS_FORCED_SHUTDOWN(mp))
4483                 return -EIO;
4484
4485         ifp = XFS_IFORK_PTR(ip, whichfork);
4486
4487         XFS_STATS_INC(mp, xs_blk_mapw);
4488
4489         if (*firstblock == NULLFSBLOCK) {
4490                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4491                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4492                 else
4493                         bma.minleft = 1;
4494         } else {
4495                 bma.minleft = 0;
4496         }
4497
4498         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4499                 error = xfs_iread_extents(tp, ip, whichfork);
4500                 if (error)
4501                         goto error0;
4502         }
4503
4504         n = 0;
4505         end = bno + len;
4506         obno = bno;
4507
4508         if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.idx, &bma.got))
4509                 eof = true;
4510         if (!xfs_iext_get_extent(ifp, bma.idx - 1, &bma.prev))
4511                 bma.prev.br_startoff = NULLFILEOFF;
4512         bma.tp = tp;
4513         bma.ip = ip;
4514         bma.total = total;
4515         bma.datatype = 0;
4516         bma.dfops = dfops;
4517         bma.firstblock = firstblock;
4518
4519         while (bno < end && n < *nmap) {
4520                 bool                    need_alloc = false, wasdelay = false;
4521
4522                 /* in hole or beyoned EOF? */
4523                 if (eof || bma.got.br_startoff > bno) {
4524                         if (flags & XFS_BMAPI_DELALLOC) {
4525                                 /*
4526                                  * For the COW fork we can reasonably get a
4527                                  * request for converting an extent that races
4528                                  * with other threads already having converted
4529                                  * part of it, as there converting COW to
4530                                  * regular blocks is not protected using the
4531                                  * IOLOCK.
4532                                  */
4533                                 ASSERT(flags & XFS_BMAPI_COWFORK);
4534                                 if (!(flags & XFS_BMAPI_COWFORK)) {
4535                                         error = -EIO;
4536                                         goto error0;
4537                                 }
4538
4539                                 if (eof || bno >= end)
4540                                         break;
4541                         } else {
4542                                 need_alloc = true;
4543                         }
4544                 } else if (isnullstartblock(bma.got.br_startblock)) {
4545                         wasdelay = true;
4546                 }
4547
4548                 /*
4549                  * First, deal with the hole before the allocated space
4550                  * that we found, if any.
4551                  */
4552                 if (need_alloc || wasdelay) {
4553                         bma.eof = eof;
4554                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4555                         bma.wasdel = wasdelay;
4556                         bma.offset = bno;
4557                         bma.flags = flags;
4558
4559                         /*
4560                          * There's a 32/64 bit type mismatch between the
4561                          * allocation length request (which can be 64 bits in
4562                          * length) and the bma length request, which is
4563                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4564                          * check for 32-bit overflows and handle them here.
4565                          */
4566                         if (len > (xfs_filblks_t)MAXEXTLEN)
4567                                 bma.length = MAXEXTLEN;
4568                         else
4569                                 bma.length = len;
4570
4571                         ASSERT(len > 0);
4572                         ASSERT(bma.length > 0);
4573                         error = xfs_bmapi_allocate(&bma);
4574                         if (error)
4575                                 goto error0;
4576                         if (bma.blkno == NULLFSBLOCK)
4577                                 break;
4578
4579                         /*
4580                          * If this is a CoW allocation, record the data in
4581                          * the refcount btree for orphan recovery.
4582                          */
4583                         if (whichfork == XFS_COW_FORK) {
4584                                 error = xfs_refcount_alloc_cow_extent(mp, dfops,
4585                                                 bma.blkno, bma.length);
4586                                 if (error)
4587                                         goto error0;
4588                         }
4589                 }
4590
4591                 /* Deal with the allocated space we found.  */
4592                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4593                                                         end, n, flags);
4594
4595                 /* Execute unwritten extent conversion if necessary */
4596                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4597                 if (error == -EAGAIN)
4598                         continue;
4599                 if (error)
4600                         goto error0;
4601
4602                 /* update the extent map to return */
4603                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4604
4605                 /*
4606                  * If we're done, stop now.  Stop when we've allocated
4607                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4608                  * the transaction may get too big.
4609                  */
4610                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4611                         break;
4612
4613                 /* Else go on to the next record. */
4614                 bma.prev = bma.got;
4615                 if (!xfs_iext_get_extent(ifp, ++bma.idx, &bma.got))
4616                         eof = true;
4617         }
4618         *nmap = n;
4619
4620         /*
4621          * Transform from btree to extents, give it cur.
4622          */
4623         if (xfs_bmap_wants_extents(ip, whichfork)) {
4624                 int             tmp_logflags = 0;
4625
4626                 ASSERT(bma.cur);
4627                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4628                         &tmp_logflags, whichfork);
4629                 bma.logflags |= tmp_logflags;
4630                 if (error)
4631                         goto error0;
4632         }
4633
4634         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4635                XFS_IFORK_NEXTENTS(ip, whichfork) >
4636                 XFS_IFORK_MAXEXT(ip, whichfork));
4637         error = 0;
4638 error0:
4639         /*
4640          * Log everything.  Do this after conversion, there's no point in
4641          * logging the extent records if we've converted to btree format.
4642          */
4643         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4644             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4645                 bma.logflags &= ~xfs_ilog_fext(whichfork);
4646         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4647                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4648                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4649         /*
4650          * Log whatever the flags say, even if error.  Otherwise we might miss
4651          * detecting a case where the data is changed, there's an error,
4652          * and it's not logged so we don't shutdown when we should.
4653          */
4654         if (bma.logflags)
4655                 xfs_trans_log_inode(tp, ip, bma.logflags);
4656
4657         if (bma.cur) {
4658                 if (!error) {
4659                         ASSERT(*firstblock == NULLFSBLOCK ||
4660                                XFS_FSB_TO_AGNO(mp, *firstblock) <=
4661                                XFS_FSB_TO_AGNO(mp,
4662                                        bma.cur->bc_private.b.firstblock));
4663                         *firstblock = bma.cur->bc_private.b.firstblock;
4664                 }
4665                 xfs_btree_del_cursor(bma.cur,
4666                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4667         }
4668         if (!error)
4669                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4670                         orig_nmap, *nmap);
4671         return error;
4672 }
4673
4674 static int
4675 xfs_bmapi_remap(
4676         struct xfs_trans        *tp,
4677         struct xfs_inode        *ip,
4678         xfs_fileoff_t           bno,
4679         xfs_filblks_t           len,
4680         xfs_fsblock_t           startblock,
4681         struct xfs_defer_ops    *dfops)
4682 {
4683         struct xfs_mount        *mp = ip->i_mount;
4684         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4685         struct xfs_btree_cur    *cur = NULL;
4686         xfs_fsblock_t           firstblock = NULLFSBLOCK;
4687         struct xfs_bmbt_irec    got;
4688         xfs_extnum_t            idx;
4689         int                     logflags = 0, error;
4690
4691         ASSERT(len > 0);
4692         ASSERT(len <= (xfs_filblks_t)MAXEXTLEN);
4693         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4694
4695         if (unlikely(XFS_TEST_ERROR(
4696             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4697              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4698              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4699                 XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW, mp);
4700                 return -EFSCORRUPTED;
4701         }
4702
4703         if (XFS_FORCED_SHUTDOWN(mp))
4704                 return -EIO;
4705
4706         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4707                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4708                 if (error)
4709                         return error;
4710         }
4711
4712         if (xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got)) {
4713                 /* make sure we only reflink into a hole. */
4714                 ASSERT(got.br_startoff > bno);
4715                 ASSERT(got.br_startoff - bno >= len);
4716         }
4717
4718         ip->i_d.di_nblocks += len;
4719         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4720
4721         if (ifp->if_flags & XFS_IFBROOT) {
4722                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
4723                 cur->bc_private.b.firstblock = firstblock;
4724                 cur->bc_private.b.dfops = dfops;
4725                 cur->bc_private.b.flags = 0;
4726         }
4727
4728         got.br_startoff = bno;
4729         got.br_startblock = startblock;
4730         got.br_blockcount = len;
4731         got.br_state = XFS_EXT_NORM;
4732
4733         error = xfs_bmap_add_extent_hole_real(tp, ip, XFS_DATA_FORK, &idx, &cur,
4734                         &got, &firstblock, dfops, &logflags);
4735         if (error)
4736                 goto error0;
4737
4738         if (xfs_bmap_wants_extents(ip, XFS_DATA_FORK)) {
4739                 int             tmp_logflags = 0;
4740
4741                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4742                         &tmp_logflags, XFS_DATA_FORK);
4743                 logflags |= tmp_logflags;
4744         }
4745
4746 error0:
4747         if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS)
4748                 logflags &= ~XFS_ILOG_DEXT;
4749         else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
4750                 logflags &= ~XFS_ILOG_DBROOT;
4751
4752         if (logflags)
4753                 xfs_trans_log_inode(tp, ip, logflags);
4754         if (cur) {
4755                 xfs_btree_del_cursor(cur,
4756                                 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4757         }
4758         return error;
4759 }
4760
4761 /*
4762  * When a delalloc extent is split (e.g., due to a hole punch), the original
4763  * indlen reservation must be shared across the two new extents that are left
4764  * behind.
4765  *
4766  * Given the original reservation and the worst case indlen for the two new
4767  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4768  * reservation fairly across the two new extents. If necessary, steal available
4769  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4770  * ores == 1). The number of stolen blocks is returned. The availability and
4771  * subsequent accounting of stolen blocks is the responsibility of the caller.
4772  */
4773 static xfs_filblks_t
4774 xfs_bmap_split_indlen(
4775         xfs_filblks_t                   ores,           /* original res. */
4776         xfs_filblks_t                   *indlen1,       /* ext1 worst indlen */
4777         xfs_filblks_t                   *indlen2,       /* ext2 worst indlen */
4778         xfs_filblks_t                   avail)          /* stealable blocks */
4779 {
4780         xfs_filblks_t                   len1 = *indlen1;
4781         xfs_filblks_t                   len2 = *indlen2;
4782         xfs_filblks_t                   nres = len1 + len2; /* new total res. */
4783         xfs_filblks_t                   stolen = 0;
4784         xfs_filblks_t                   resfactor;
4785
4786         /*
4787          * Steal as many blocks as we can to try and satisfy the worst case
4788          * indlen for both new extents.
4789          */
4790         if (ores < nres && avail)
4791                 stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4792         ores += stolen;
4793
4794          /* nothing else to do if we've satisfied the new reservation */
4795         if (ores >= nres)
4796                 return stolen;
4797
4798         /*
4799          * We can't meet the total required reservation for the two extents.
4800          * Calculate the percent of the overall shortage between both extents
4801          * and apply this percentage to each of the requested indlen values.
4802          * This distributes the shortage fairly and reduces the chances that one
4803          * of the two extents is left with nothing when extents are repeatedly
4804          * split.
4805          */
4806         resfactor = (ores * 100);
4807         do_div(resfactor, nres);
4808         len1 *= resfactor;
4809         do_div(len1, 100);
4810         len2 *= resfactor;
4811         do_div(len2, 100);
4812         ASSERT(len1 + len2 <= ores);
4813         ASSERT(len1 < *indlen1 && len2 < *indlen2);
4814
4815         /*
4816          * Hand out the remainder to each extent. If one of the two reservations
4817          * is zero, we want to make sure that one gets a block first. The loop
4818          * below starts with len1, so hand len2 a block right off the bat if it
4819          * is zero.
4820          */
4821         ores -= (len1 + len2);
4822         ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4823         if (ores && !len2 && *indlen2) {
4824                 len2++;
4825                 ores--;
4826         }
4827         while (ores) {
4828                 if (len1 < *indlen1) {
4829                         len1++;
4830                         ores--;
4831                 }
4832                 if (!ores)
4833                         break;
4834                 if (len2 < *indlen2) {
4835                         len2++;
4836                         ores--;
4837                 }
4838         }
4839
4840         *indlen1 = len1;
4841         *indlen2 = len2;
4842
4843         return stolen;
4844 }
4845
4846 int
4847 xfs_bmap_del_extent_delay(
4848         struct xfs_inode        *ip,
4849         int                     whichfork,
4850         xfs_extnum_t            *idx,
4851         struct xfs_bmbt_irec    *got,
4852         struct xfs_bmbt_irec    *del)
4853 {
4854         struct xfs_mount        *mp = ip->i_mount;
4855         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4856         struct xfs_bmbt_irec    new;
4857         int64_t                 da_old, da_new, da_diff = 0;
4858         xfs_fileoff_t           del_endoff, got_endoff;
4859         xfs_filblks_t           got_indlen, new_indlen, stolen;
4860         int                     error = 0, state = 0;
4861         bool                    isrt;
4862
4863         XFS_STATS_INC(mp, xs_del_exlist);
4864
4865         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4866         del_endoff = del->br_startoff + del->br_blockcount;
4867         got_endoff = got->br_startoff + got->br_blockcount;
4868         da_old = startblockval(got->br_startblock);
4869         da_new = 0;
4870
4871         ASSERT(*idx >= 0);
4872         ASSERT(*idx <= xfs_iext_count(ifp));
4873         ASSERT(del->br_blockcount > 0);
4874         ASSERT(got->br_startoff <= del->br_startoff);
4875         ASSERT(got_endoff >= del_endoff);
4876
4877         if (isrt) {
4878                 uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4879
4880                 do_div(rtexts, mp->m_sb.sb_rextsize);
4881                 xfs_mod_frextents(mp, rtexts);
4882         }
4883
4884         /*
4885          * Update the inode delalloc counter now and wait to update the
4886          * sb counters as we might have to borrow some blocks for the
4887          * indirect block accounting.
4888          */
4889         error = xfs_trans_reserve_quota_nblks(NULL, ip,
4890                         -((long)del->br_blockcount), 0,
4891                         isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4892         if (error)
4893                 return error;
4894         ip->i_delayed_blks -= del->br_blockcount;
4895
4896         if (whichfork == XFS_COW_FORK)
4897                 state |= BMAP_COWFORK;
4898
4899         if (got->br_startoff == del->br_startoff)
4900                 state |= BMAP_LEFT_CONTIG;
4901         if (got_endoff == del_endoff)
4902                 state |= BMAP_RIGHT_CONTIG;
4903
4904         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
4905         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
4906                 /*
4907                  * Matches the whole extent.  Delete the entry.
4908                  */
4909                 xfs_iext_remove(ip, *idx, 1, state);
4910                 --*idx;
4911                 break;
4912         case BMAP_LEFT_CONTIG:
4913                 /*
4914                  * Deleting the first part of the extent.
4915                  */
4916                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4917                 got->br_startoff = del_endoff;
4918                 got->br_blockcount -= del->br_blockcount;
4919                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4920                                 got->br_blockcount), da_old);
4921                 got->br_startblock = nullstartblock((int)da_new);
4922                 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
4923                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4924                 break;
4925         case BMAP_RIGHT_CONTIG:
4926                 /*
4927                  * Deleting the last part of the extent.
4928                  */
4929                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4930                 got->br_blockcount = got->br_blockcount - del->br_blockcount;
4931                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4932                                 got->br_blockcount), da_old);
4933                 got->br_startblock = nullstartblock((int)da_new);
4934                 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
4935                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4936                 break;
4937         case 0:
4938                 /*
4939                  * Deleting the middle of the extent.
4940                  *
4941                  * Distribute the original indlen reservation across the two new
4942                  * extents.  Steal blocks from the deleted extent if necessary.
4943                  * Stealing blocks simply fudges the fdblocks accounting below.
4944                  * Warn if either of the new indlen reservations is zero as this
4945                  * can lead to delalloc problems.
4946                  */
4947                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4948
4949                 got->br_blockcount = del->br_startoff - got->br_startoff;
4950                 got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4951
4952                 new.br_blockcount = got_endoff - del_endoff;
4953                 new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4954
4955                 WARN_ON_ONCE(!got_indlen || !new_indlen);
4956                 stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4957                                                        del->br_blockcount);
4958
4959                 got->br_startblock = nullstartblock((int)got_indlen);
4960                 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
4961                 trace_xfs_bmap_post_update(ip, *idx, 0, _THIS_IP_);
4962
4963                 new.br_startoff = del_endoff;
4964                 new.br_state = got->br_state;
4965                 new.br_startblock = nullstartblock((int)new_indlen);
4966
4967                 ++*idx;
4968                 xfs_iext_insert(ip, *idx, 1, &new, state);
4969
4970                 da_new = got_indlen + new_indlen - stolen;
4971                 del->br_blockcount -= stolen;
4972                 break;
4973         }
4974
4975         ASSERT(da_old >= da_new);
4976         da_diff = da_old - da_new;
4977         if (!isrt)
4978                 da_diff += del->br_blockcount;
4979         if (da_diff)
4980                 xfs_mod_fdblocks(mp, da_diff, false);
4981         return error;
4982 }
4983
4984 void
4985 xfs_bmap_del_extent_cow(
4986         struct xfs_inode        *ip,
4987         xfs_extnum_t            *idx,
4988         struct xfs_bmbt_irec    *got,
4989         struct xfs_bmbt_irec    *del)
4990 {
4991         struct xfs_mount        *mp = ip->i_mount;
4992         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
4993         struct xfs_bmbt_irec    new;
4994         xfs_fileoff_t           del_endoff, got_endoff;
4995         int                     state = BMAP_COWFORK;
4996
4997         XFS_STATS_INC(mp, xs_del_exlist);
4998
4999         del_endoff = del->br_startoff + del->br_blockcount;
5000         got_endoff = got->br_startoff + got->br_blockcount;
5001
5002         ASSERT(*idx >= 0);
5003         ASSERT(*idx <= xfs_iext_count(ifp));
5004         ASSERT(del->br_blockcount > 0);
5005         ASSERT(got->br_startoff <= del->br_startoff);
5006         ASSERT(got_endoff >= del_endoff);
5007         ASSERT(!isnullstartblock(got->br_startblock));
5008
5009         if (got->br_startoff == del->br_startoff)
5010                 state |= BMAP_LEFT_CONTIG;
5011         if (got_endoff == del_endoff)
5012                 state |= BMAP_RIGHT_CONTIG;
5013
5014         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
5015         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
5016                 /*
5017                  * Matches the whole extent.  Delete the entry.
5018                  */
5019                 xfs_iext_remove(ip, *idx, 1, state);
5020                 --*idx;
5021                 break;
5022         case BMAP_LEFT_CONTIG:
5023                 /*
5024                  * Deleting the first part of the extent.
5025                  */
5026                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5027                 got->br_startoff = del_endoff;
5028                 got->br_blockcount -= del->br_blockcount;
5029                 got->br_startblock = del->br_startblock + del->br_blockcount;
5030                 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
5031                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5032                 break;
5033         case BMAP_RIGHT_CONTIG:
5034                 /*
5035                  * Deleting the last part of the extent.
5036                  */
5037                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5038                 got->br_blockcount -= del->br_blockcount;
5039                 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
5040                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5041                 break;
5042         case 0:
5043                 /*
5044                  * Deleting the middle of the extent.
5045                  */
5046                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5047                 got->br_blockcount = del->br_startoff - got->br_startoff;
5048                 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
5049                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5050
5051                 new.br_startoff = del_endoff;
5052                 new.br_blockcount = got_endoff - del_endoff;
5053                 new.br_state = got->br_state;
5054                 new.br_startblock = del->br_startblock + del->br_blockcount;
5055
5056                 ++*idx;
5057                 xfs_iext_insert(ip, *idx, 1, &new, state);
5058                 break;
5059         }
5060 }
5061
5062 /*
5063  * Called by xfs_bmapi to update file extent records and the btree
5064  * after removing space (or undoing a delayed allocation).
5065  */
5066 STATIC int                              /* error */
5067 xfs_bmap_del_extent(
5068         xfs_inode_t             *ip,    /* incore inode pointer */
5069         xfs_trans_t             *tp,    /* current transaction pointer */
5070         xfs_extnum_t            *idx,   /* extent number to update/delete */
5071         struct xfs_defer_ops    *dfops, /* list of extents to be freed */
5072         xfs_btree_cur_t         *cur,   /* if null, not a btree */
5073         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
5074         int                     *logflagsp, /* inode logging flags */
5075         int                     whichfork, /* data or attr fork */
5076         int                     bflags) /* bmapi flags */
5077 {
5078         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
5079         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
5080         xfs_fsblock_t           del_endblock=0; /* first block past del */
5081         xfs_fileoff_t           del_endoff;     /* first offset past del */
5082         int                     delay;  /* current block is delayed allocated */
5083         int                     do_fx;  /* free extent at end of routine */
5084         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
5085         int                     error;  /* error return value */
5086         int                     flags;  /* inode logging flags */
5087         xfs_bmbt_irec_t         got;    /* current extent entry */
5088         xfs_fileoff_t           got_endoff;     /* first offset past got */
5089         int                     i;      /* temp state */
5090         xfs_ifork_t             *ifp;   /* inode fork pointer */
5091         xfs_mount_t             *mp;    /* mount structure */
5092         xfs_filblks_t           nblks;  /* quota/sb block count */
5093         xfs_bmbt_irec_t         new;    /* new record to be inserted */
5094         /* REFERENCED */
5095         uint                    qfield; /* quota field to update */
5096         xfs_filblks_t           temp;   /* for indirect length calculations */
5097         xfs_filblks_t           temp2;  /* for indirect length calculations */
5098         int                     state = 0;
5099
5100         mp = ip->i_mount;
5101         XFS_STATS_INC(mp, xs_del_exlist);
5102
5103         if (whichfork == XFS_ATTR_FORK)
5104                 state |= BMAP_ATTRFORK;
5105         else if (whichfork == XFS_COW_FORK)
5106                 state |= BMAP_COWFORK;
5107
5108         ifp = XFS_IFORK_PTR(ip, whichfork);
5109         ASSERT((*idx >= 0) && (*idx < xfs_iext_count(ifp)));
5110         ASSERT(del->br_blockcount > 0);
5111         ep = xfs_iext_get_ext(ifp, *idx);
5112         xfs_bmbt_get_all(ep, &got);
5113         ASSERT(got.br_startoff <= del->br_startoff);
5114         del_endoff = del->br_startoff + del->br_blockcount;
5115         got_endoff = got.br_startoff + got.br_blockcount;
5116         ASSERT(got_endoff >= del_endoff);
5117         delay = isnullstartblock(got.br_startblock);
5118         ASSERT(isnullstartblock(del->br_startblock) == delay);
5119         flags = 0;
5120         qfield = 0;
5121         error = 0;
5122         /*
5123          * If deleting a real allocation, must free up the disk space.
5124          */
5125         if (!delay) {
5126                 flags = XFS_ILOG_CORE;
5127                 /*
5128                  * Realtime allocation.  Free it and record di_nblocks update.
5129                  */
5130                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5131                         xfs_fsblock_t   bno;
5132                         xfs_filblks_t   len;
5133
5134                         ASSERT(do_mod(del->br_blockcount,
5135                                       mp->m_sb.sb_rextsize) == 0);
5136                         ASSERT(do_mod(del->br_startblock,
5137                                       mp->m_sb.sb_rextsize) == 0);
5138                         bno = del->br_startblock;
5139                         len = del->br_blockcount;
5140                         do_div(bno, mp->m_sb.sb_rextsize);
5141                         do_div(len, mp->m_sb.sb_rextsize);
5142                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5143                         if (error)
5144                                 goto done;
5145                         do_fx = 0;
5146                         nblks = len * mp->m_sb.sb_rextsize;
5147                         qfield = XFS_TRANS_DQ_RTBCOUNT;
5148                 }
5149                 /*
5150                  * Ordinary allocation.
5151                  */
5152                 else {
5153                         do_fx = 1;
5154                         nblks = del->br_blockcount;
5155                         qfield = XFS_TRANS_DQ_BCOUNT;
5156                 }
5157                 /*
5158                  * Set up del_endblock and cur for later.
5159                  */
5160                 del_endblock = del->br_startblock + del->br_blockcount;
5161                 if (cur) {
5162                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5163                                         got.br_startblock, got.br_blockcount,
5164                                         &i)))
5165                                 goto done;
5166                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5167                 }
5168                 da_old = da_new = 0;
5169         } else {
5170                 da_old = startblockval(got.br_startblock);
5171                 da_new = 0;
5172                 nblks = 0;
5173                 do_fx = 0;
5174         }
5175
5176         /*
5177          * Set flag value to use in switch statement.
5178          * Left-contig is 2, right-contig is 1.
5179          */
5180         switch (((got.br_startoff == del->br_startoff) << 1) |
5181                 (got_endoff == del_endoff)) {
5182         case 3:
5183                 /*
5184                  * Matches the whole extent.  Delete the entry.
5185                  */
5186                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5187                 xfs_iext_remove(ip, *idx, 1,
5188                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
5189                 --*idx;
5190                 if (delay)
5191                         break;
5192
5193                 XFS_IFORK_NEXT_SET(ip, whichfork,
5194                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5195                 flags |= XFS_ILOG_CORE;
5196                 if (!cur) {
5197                         flags |= xfs_ilog_fext(whichfork);
5198                         break;
5199                 }
5200                 if ((error = xfs_btree_delete(cur, &i)))
5201                         goto done;
5202                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5203                 break;
5204
5205         case 2:
5206                 /*
5207                  * Deleting the first part of the extent.
5208                  */
5209                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5210                 xfs_bmbt_set_startoff(ep, del_endoff);
5211                 temp = got.br_blockcount - del->br_blockcount;
5212                 xfs_bmbt_set_blockcount(ep, temp);
5213                 if (delay) {
5214                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5215                                 da_old);
5216                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5217                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5218                         da_new = temp;
5219                         break;
5220                 }
5221                 xfs_bmbt_set_startblock(ep, del_endblock);
5222                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5223                 if (!cur) {
5224                         flags |= xfs_ilog_fext(whichfork);
5225                         break;
5226                 }
5227                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
5228                                 got.br_blockcount - del->br_blockcount,
5229                                 got.br_state)))
5230                         goto done;
5231                 break;
5232
5233         case 1:
5234                 /*
5235                  * Deleting the last part of the extent.
5236                  */
5237                 temp = got.br_blockcount - del->br_blockcount;
5238                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5239                 xfs_bmbt_set_blockcount(ep, temp);
5240                 if (delay) {
5241                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5242                                 da_old);
5243                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5244                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5245                         da_new = temp;
5246                         break;
5247                 }
5248                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5249                 if (!cur) {
5250                         flags |= xfs_ilog_fext(whichfork);
5251                         break;
5252                 }
5253                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
5254                                 got.br_startblock,
5255                                 got.br_blockcount - del->br_blockcount,
5256                                 got.br_state)))
5257                         goto done;
5258                 break;
5259
5260         case 0:
5261                 /*
5262                  * Deleting the middle of the extent.
5263                  */
5264                 temp = del->br_startoff - got.br_startoff;
5265                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5266                 xfs_bmbt_set_blockcount(ep, temp);
5267                 new.br_startoff = del_endoff;
5268                 temp2 = got_endoff - del_endoff;
5269                 new.br_blockcount = temp2;
5270                 new.br_state = got.br_state;
5271                 if (!delay) {
5272                         new.br_startblock = del_endblock;
5273                         flags |= XFS_ILOG_CORE;
5274                         if (cur) {
5275                                 if ((error = xfs_bmbt_update(cur,
5276                                                 got.br_startoff,
5277                                                 got.br_startblock, temp,
5278                                                 got.br_state)))
5279                                         goto done;
5280                                 if ((error = xfs_btree_increment(cur, 0, &i)))
5281                                         goto done;
5282                                 cur->bc_rec.b = new;
5283                                 error = xfs_btree_insert(cur, &i);
5284                                 if (error && error != -ENOSPC)
5285                                         goto done;
5286                                 /*
5287                                  * If get no-space back from btree insert,
5288                                  * it tried a split, and we have a zero
5289                                  * block reservation.
5290                                  * Fix up our state and return the error.
5291                                  */
5292                                 if (error == -ENOSPC) {
5293                                         /*
5294                                          * Reset the cursor, don't trust
5295                                          * it after any insert operation.
5296                                          */
5297                                         if ((error = xfs_bmbt_lookup_eq(cur,
5298                                                         got.br_startoff,
5299                                                         got.br_startblock,
5300                                                         temp, &i)))
5301                                                 goto done;
5302                                         XFS_WANT_CORRUPTED_GOTO(mp,
5303                                                                 i == 1, done);
5304                                         /*
5305                                          * Update the btree record back
5306                                          * to the original value.
5307                                          */
5308                                         if ((error = xfs_bmbt_update(cur,
5309                                                         got.br_startoff,
5310                                                         got.br_startblock,
5311                                                         got.br_blockcount,
5312                                                         got.br_state)))
5313                                                 goto done;
5314                                         /*
5315                                          * Reset the extent record back
5316                                          * to the original value.
5317                                          */
5318                                         xfs_bmbt_set_blockcount(ep,
5319                                                 got.br_blockcount);
5320                                         flags = 0;
5321                                         error = -ENOSPC;
5322                                         goto done;
5323                                 }
5324                                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5325                         } else
5326                                 flags |= xfs_ilog_fext(whichfork);
5327                         XFS_IFORK_NEXT_SET(ip, whichfork,
5328                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5329                 } else {
5330                         xfs_filblks_t   stolen;
5331                         ASSERT(whichfork == XFS_DATA_FORK);
5332
5333                         /*
5334                          * Distribute the original indlen reservation across the
5335                          * two new extents. Steal blocks from the deleted extent
5336                          * if necessary. Stealing blocks simply fudges the
5337                          * fdblocks accounting in xfs_bunmapi().
5338                          */
5339                         temp = xfs_bmap_worst_indlen(ip, got.br_blockcount);
5340                         temp2 = xfs_bmap_worst_indlen(ip, new.br_blockcount);
5341                         stolen = xfs_bmap_split_indlen(da_old, &temp, &temp2,
5342                                                        del->br_blockcount);
5343                         da_new = temp + temp2 - stolen;
5344                         del->br_blockcount -= stolen;
5345
5346                         /*
5347                          * Set the reservation for each extent. Warn if either
5348                          * is zero as this can lead to delalloc problems.
5349                          */
5350                         WARN_ON_ONCE(!temp || !temp2);
5351                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5352                         new.br_startblock = nullstartblock((int)temp2);
5353                 }
5354                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5355                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5356                 ++*idx;
5357                 break;
5358         }
5359
5360         /* remove reverse mapping */
5361         if (!delay) {
5362                 error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, del);
5363                 if (error)
5364                         goto done;
5365         }
5366
5367         /*
5368          * If we need to, add to list of extents to delete.
5369          */
5370         if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5371                 if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5372                         error = xfs_refcount_decrease_extent(mp, dfops, del);
5373                         if (error)
5374                                 goto done;
5375                 } else
5376                         xfs_bmap_add_free(mp, dfops, del->br_startblock,
5377                                         del->br_blockcount, NULL);
5378         }
5379
5380         /*
5381          * Adjust inode # blocks in the file.
5382          */
5383         if (nblks)
5384                 ip->i_d.di_nblocks -= nblks;
5385         /*
5386          * Adjust quota data.
5387          */
5388         if (qfield && !(bflags & XFS_BMAPI_REMAP))
5389                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5390
5391         /*
5392          * Account for change in delayed indirect blocks.
5393          * Nothing to do for disk quota accounting here.
5394          */
5395         ASSERT(da_old >= da_new);
5396         if (da_old > da_new)
5397                 xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false);
5398 done:
5399         *logflagsp = flags;
5400         return error;
5401 }
5402
5403 /*
5404  * Unmap (remove) blocks from a file.
5405  * If nexts is nonzero then the number of extents to remove is limited to
5406  * that value.  If not all extents in the block range can be removed then
5407  * *done is set.
5408  */
5409 int                                             /* error */
5410 __xfs_bunmapi(
5411         xfs_trans_t             *tp,            /* transaction pointer */
5412         struct xfs_inode        *ip,            /* incore inode */
5413         xfs_fileoff_t           bno,            /* starting offset to unmap */
5414         xfs_filblks_t           *rlen,          /* i/o: amount remaining */
5415         int                     flags,          /* misc flags */
5416         xfs_extnum_t            nexts,          /* number of extents max */
5417         xfs_fsblock_t           *firstblock,    /* first allocated block
5418                                                    controls a.g. for allocs */
5419         struct xfs_defer_ops    *dfops)         /* i/o: deferred updates */
5420 {
5421         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5422         xfs_bmbt_irec_t         del;            /* extent being deleted */
5423         int                     error;          /* error return value */
5424         xfs_extnum_t            extno;          /* extent number in list */
5425         xfs_bmbt_irec_t         got;            /* current extent record */
5426         xfs_ifork_t             *ifp;           /* inode fork pointer */
5427         int                     isrt;           /* freeing in rt area */
5428         xfs_extnum_t            lastx;          /* last extent index used */
5429         int                     logflags;       /* transaction logging flags */
5430         xfs_extlen_t            mod;            /* rt extent offset */
5431         xfs_mount_t             *mp;            /* mount structure */
5432         xfs_fileoff_t           start;          /* first file offset deleted */
5433         int                     tmp_logflags;   /* partial logging flags */
5434         int                     wasdel;         /* was a delayed alloc extent */
5435         int                     whichfork;      /* data or attribute fork */
5436         xfs_fsblock_t           sum;
5437         xfs_filblks_t           len = *rlen;    /* length to unmap in file */
5438
5439         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5440
5441         whichfork = xfs_bmapi_whichfork(flags);
5442         ASSERT(whichfork != XFS_COW_FORK);
5443         ifp = XFS_IFORK_PTR(ip, whichfork);
5444         if (unlikely(
5445             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5446             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5447                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5448                                  ip->i_mount);
5449                 return -EFSCORRUPTED;
5450         }
5451         mp = ip->i_mount;
5452         if (XFS_FORCED_SHUTDOWN(mp))
5453                 return -EIO;
5454
5455         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5456         ASSERT(len > 0);
5457         ASSERT(nexts >= 0);
5458
5459         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5460             (error = xfs_iread_extents(tp, ip, whichfork)))
5461                 return error;
5462         if (xfs_iext_count(ifp) == 0) {
5463                 *rlen = 0;
5464                 return 0;
5465         }
5466         XFS_STATS_INC(mp, xs_blk_unmap);
5467         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5468         start = bno;
5469         bno = start + len - 1;
5470
5471         /*
5472          * Check to see if the given block number is past the end of the
5473          * file, back up to the last block if so...
5474          */
5475         if (!xfs_iext_lookup_extent(ip, ifp, bno, &lastx, &got)) {
5476                 ASSERT(lastx > 0);
5477                 xfs_iext_get_extent(ifp, --lastx, &got);
5478                 bno = got.br_startoff + got.br_blockcount - 1;
5479         }
5480
5481         logflags = 0;
5482         if (ifp->if_flags & XFS_IFBROOT) {
5483                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5484                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5485                 cur->bc_private.b.firstblock = *firstblock;
5486                 cur->bc_private.b.dfops = dfops;
5487                 cur->bc_private.b.flags = 0;
5488         } else
5489                 cur = NULL;
5490
5491         if (isrt) {
5492                 /*
5493                  * Synchronize by locking the bitmap inode.
5494                  */
5495                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5496                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5497                 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5498                 xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5499         }
5500
5501         extno = 0;
5502         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5503                (nexts == 0 || extno < nexts)) {
5504                 /*
5505                  * Is the found extent after a hole in which bno lives?
5506                  * Just back up to the previous extent, if so.
5507                  */
5508                 if (got.br_startoff > bno) {
5509                         if (--lastx < 0)
5510                                 break;
5511                         xfs_iext_get_extent(ifp, lastx, &got);
5512                 }
5513                 /*
5514                  * Is the last block of this extent before the range
5515                  * we're supposed to delete?  If so, we're done.
5516                  */
5517                 bno = XFS_FILEOFF_MIN(bno,
5518                         got.br_startoff + got.br_blockcount - 1);
5519                 if (bno < start)
5520                         break;
5521                 /*
5522                  * Then deal with the (possibly delayed) allocated space
5523                  * we found.
5524                  */
5525                 del = got;
5526                 wasdel = isnullstartblock(del.br_startblock);
5527                 if (got.br_startoff < start) {
5528                         del.br_startoff = start;
5529                         del.br_blockcount -= start - got.br_startoff;
5530                         if (!wasdel)
5531                                 del.br_startblock += start - got.br_startoff;
5532                 }
5533                 if (del.br_startoff + del.br_blockcount > bno + 1)
5534                         del.br_blockcount = bno + 1 - del.br_startoff;
5535                 sum = del.br_startblock + del.br_blockcount;
5536                 if (isrt &&
5537                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5538                         /*
5539                          * Realtime extent not lined up at the end.
5540                          * The extent could have been split into written
5541                          * and unwritten pieces, or we could just be
5542                          * unmapping part of it.  But we can't really
5543                          * get rid of part of a realtime extent.
5544                          */
5545                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5546                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5547                                 /*
5548                                  * This piece is unwritten, or we're not
5549                                  * using unwritten extents.  Skip over it.
5550                                  */
5551                                 ASSERT(bno >= mod);
5552                                 bno -= mod > del.br_blockcount ?
5553                                         del.br_blockcount : mod;
5554                                 if (bno < got.br_startoff) {
5555                                         if (--lastx >= 0)
5556                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5557                                                         ifp, lastx), &got);
5558                                 }
5559                                 continue;
5560                         }
5561                         /*
5562                          * It's written, turn it unwritten.
5563                          * This is better than zeroing it.
5564                          */
5565                         ASSERT(del.br_state == XFS_EXT_NORM);
5566                         ASSERT(tp->t_blk_res > 0);
5567                         /*
5568                          * If this spans a realtime extent boundary,
5569                          * chop it back to the start of the one we end at.
5570                          */
5571                         if (del.br_blockcount > mod) {
5572                                 del.br_startoff += del.br_blockcount - mod;
5573                                 del.br_startblock += del.br_blockcount - mod;
5574                                 del.br_blockcount = mod;
5575                         }
5576                         del.br_state = XFS_EXT_UNWRITTEN;
5577                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5578                                         whichfork, &lastx, &cur, &del,
5579                                         firstblock, dfops, &logflags);
5580                         if (error)
5581                                 goto error0;
5582                         goto nodelete;
5583                 }
5584                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5585                         /*
5586                          * Realtime extent is lined up at the end but not
5587                          * at the front.  We'll get rid of full extents if
5588                          * we can.
5589                          */
5590                         mod = mp->m_sb.sb_rextsize - mod;
5591                         if (del.br_blockcount > mod) {
5592                                 del.br_blockcount -= mod;
5593                                 del.br_startoff += mod;
5594                                 del.br_startblock += mod;
5595                         } else if ((del.br_startoff == start &&
5596                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5597                                      tp->t_blk_res == 0)) ||
5598                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5599                                 /*
5600                                  * Can't make it unwritten.  There isn't
5601                                  * a full extent here so just skip it.
5602                                  */
5603                                 ASSERT(bno >= del.br_blockcount);
5604                                 bno -= del.br_blockcount;
5605                                 if (got.br_startoff > bno && --lastx >= 0)
5606                                         xfs_iext_get_extent(ifp, lastx, &got);
5607                                 continue;
5608                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5609                                 struct xfs_bmbt_irec    prev;
5610
5611                                 /*
5612                                  * This one is already unwritten.
5613                                  * It must have a written left neighbor.
5614                                  * Unwrite the killed part of that one and
5615                                  * try again.
5616                                  */
5617                                 ASSERT(lastx > 0);
5618                                 xfs_iext_get_extent(ifp, lastx - 1, &prev);
5619                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5620                                 ASSERT(!isnullstartblock(prev.br_startblock));
5621                                 ASSERT(del.br_startblock ==
5622                                        prev.br_startblock + prev.br_blockcount);
5623                                 if (prev.br_startoff < start) {
5624                                         mod = start - prev.br_startoff;
5625                                         prev.br_blockcount -= mod;
5626                                         prev.br_startblock += mod;
5627                                         prev.br_startoff = start;
5628                                 }
5629                                 prev.br_state = XFS_EXT_UNWRITTEN;
5630                                 lastx--;
5631                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5632                                                 ip, whichfork, &lastx, &cur,
5633                                                 &prev, firstblock, dfops,
5634                                                 &logflags);
5635                                 if (error)
5636                                         goto error0;
5637                                 goto nodelete;
5638                         } else {
5639                                 ASSERT(del.br_state == XFS_EXT_NORM);
5640                                 del.br_state = XFS_EXT_UNWRITTEN;
5641                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5642                                                 ip, whichfork, &lastx, &cur,
5643                                                 &del, firstblock, dfops,
5644                                                 &logflags);
5645                                 if (error)
5646                                         goto error0;
5647                                 goto nodelete;
5648                         }
5649                 }
5650
5651                 /*
5652                  * If it's the case where the directory code is running
5653                  * with no block reservation, and the deleted block is in
5654                  * the middle of its extent, and the resulting insert
5655                  * of an extent would cause transformation to btree format,
5656                  * then reject it.  The calling code will then swap
5657                  * blocks around instead.
5658                  * We have to do this now, rather than waiting for the
5659                  * conversion to btree format, since the transaction
5660                  * will be dirty.
5661                  */
5662                 if (!wasdel && tp->t_blk_res == 0 &&
5663                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5664                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5665                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5666                     del.br_startoff > got.br_startoff &&
5667                     del.br_startoff + del.br_blockcount <
5668                     got.br_startoff + got.br_blockcount) {
5669                         error = -ENOSPC;
5670                         goto error0;
5671                 }
5672
5673                 /*
5674                  * Unreserve quota and update realtime free space, if
5675                  * appropriate. If delayed allocation, update the inode delalloc
5676                  * counter now and wait to update the sb counters as
5677                  * xfs_bmap_del_extent() might need to borrow some blocks.
5678                  */
5679                 if (wasdel) {
5680                         ASSERT(startblockval(del.br_startblock) > 0);
5681                         if (isrt) {
5682                                 xfs_filblks_t rtexts;
5683
5684                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5685                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5686                                 xfs_mod_frextents(mp, (int64_t)rtexts);
5687                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5688                                         ip, -((long)del.br_blockcount), 0,
5689                                         XFS_QMOPT_RES_RTBLKS);
5690                         } else {
5691                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5692                                         ip, -((long)del.br_blockcount), 0,
5693                                         XFS_QMOPT_RES_REGBLKS);
5694                         }
5695                         ip->i_delayed_blks -= del.br_blockcount;
5696                         if (cur)
5697                                 cur->bc_private.b.flags |=
5698                                         XFS_BTCUR_BPRV_WASDEL;
5699                 } else if (cur)
5700                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5701
5702                 error = xfs_bmap_del_extent(ip, tp, &lastx, dfops, cur, &del,
5703                                 &tmp_logflags, whichfork, flags);
5704                 logflags |= tmp_logflags;
5705                 if (error)
5706                         goto error0;
5707
5708                 if (!isrt && wasdel)
5709                         xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount, false);
5710
5711                 bno = del.br_startoff - 1;
5712 nodelete:
5713                 /*
5714                  * If not done go on to the next (previous) record.
5715                  */
5716                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5717                         if (lastx >= 0) {
5718                                 xfs_iext_get_extent(ifp, lastx, &got);
5719                                 if (got.br_startoff > bno && --lastx >= 0)
5720                                         xfs_iext_get_extent(ifp, lastx, &got);
5721                         }
5722                         extno++;
5723                 }
5724         }
5725         if (bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0)
5726                 *rlen = 0;
5727         else
5728                 *rlen = bno - start + 1;
5729
5730         /*
5731          * Convert to a btree if necessary.
5732          */
5733         if (xfs_bmap_needs_btree(ip, whichfork)) {
5734                 ASSERT(cur == NULL);
5735                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops,
5736                         &cur, 0, &tmp_logflags, whichfork);
5737                 logflags |= tmp_logflags;
5738                 if (error)
5739                         goto error0;
5740         }
5741         /*
5742          * transform from btree to extents, give it cur
5743          */
5744         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5745                 ASSERT(cur != NULL);
5746                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5747                         whichfork);
5748                 logflags |= tmp_logflags;
5749                 if (error)
5750                         goto error0;
5751         }
5752         /*
5753          * transform from extents to local?
5754          */
5755         error = 0;
5756 error0:
5757         /*
5758          * Log everything.  Do this after conversion, there's no point in
5759          * logging the extent records if we've converted to btree format.
5760          */
5761         if ((logflags & xfs_ilog_fext(whichfork)) &&
5762             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5763                 logflags &= ~xfs_ilog_fext(whichfork);
5764         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5765                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5766                 logflags &= ~xfs_ilog_fbroot(whichfork);
5767         /*
5768          * Log inode even in the error case, if the transaction
5769          * is dirty we'll need to shut down the filesystem.
5770          */
5771         if (logflags)
5772                 xfs_trans_log_inode(tp, ip, logflags);
5773         if (cur) {
5774                 if (!error) {
5775                         *firstblock = cur->bc_private.b.firstblock;
5776                         cur->bc_private.b.allocated = 0;
5777                 }
5778                 xfs_btree_del_cursor(cur,
5779                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5780         }
5781         return error;
5782 }
5783
5784 /* Unmap a range of a file. */
5785 int
5786 xfs_bunmapi(
5787         xfs_trans_t             *tp,
5788         struct xfs_inode        *ip,
5789         xfs_fileoff_t           bno,
5790         xfs_filblks_t           len,
5791         int                     flags,
5792         xfs_extnum_t            nexts,
5793         xfs_fsblock_t           *firstblock,
5794         struct xfs_defer_ops    *dfops,
5795         int                     *done)
5796 {
5797         int                     error;
5798
5799         error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts, firstblock,
5800                         dfops);
5801         *done = (len == 0);
5802         return error;
5803 }
5804
5805 /*
5806  * Determine whether an extent shift can be accomplished by a merge with the
5807  * extent that precedes the target hole of the shift.
5808  */
5809 STATIC bool
5810 xfs_bmse_can_merge(
5811         struct xfs_bmbt_irec    *left,  /* preceding extent */
5812         struct xfs_bmbt_irec    *got,   /* current extent to shift */
5813         xfs_fileoff_t           shift)  /* shift fsb */
5814 {
5815         xfs_fileoff_t           startoff;
5816
5817         startoff = got->br_startoff - shift;
5818
5819         /*
5820          * The extent, once shifted, must be adjacent in-file and on-disk with
5821          * the preceding extent.
5822          */
5823         if ((left->br_startoff + left->br_blockcount != startoff) ||
5824             (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5825             (left->br_state != got->br_state) ||
5826             (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5827                 return false;
5828
5829         return true;
5830 }
5831
5832 /*
5833  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5834  * hole in the file. If an extent shift would result in the extent being fully
5835  * adjacent to the extent that currently precedes the hole, we can merge with
5836  * the preceding extent rather than do the shift.
5837  *
5838  * This function assumes the caller has verified a shift-by-merge is possible
5839  * with the provided extents via xfs_bmse_can_merge().
5840  */
5841 STATIC int
5842 xfs_bmse_merge(
5843         struct xfs_inode                *ip,
5844         int                             whichfork,
5845         xfs_fileoff_t                   shift,          /* shift fsb */
5846         int                             current_ext,    /* idx of gotp */
5847         struct xfs_bmbt_rec_host        *gotp,          /* extent to shift */
5848         struct xfs_bmbt_rec_host        *leftp,         /* preceding extent */
5849         struct xfs_btree_cur            *cur,
5850         int                             *logflags)      /* output */
5851 {
5852         struct xfs_bmbt_irec            got;
5853         struct xfs_bmbt_irec            left;
5854         xfs_filblks_t                   blockcount;
5855         int                             error, i;
5856         struct xfs_mount                *mp = ip->i_mount;
5857
5858         xfs_bmbt_get_all(gotp, &got);
5859         xfs_bmbt_get_all(leftp, &left);
5860         blockcount = left.br_blockcount + got.br_blockcount;
5861
5862         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5863         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5864         ASSERT(xfs_bmse_can_merge(&left, &got, shift));
5865
5866         /*
5867          * Merge the in-core extents. Note that the host record pointers and
5868          * current_ext index are invalid once the extent has been removed via
5869          * xfs_iext_remove().
5870          */
5871         xfs_bmbt_set_blockcount(leftp, blockcount);
5872         xfs_iext_remove(ip, current_ext, 1, 0);
5873
5874         /*
5875          * Update the on-disk extent count, the btree if necessary and log the
5876          * inode.
5877          */
5878         XFS_IFORK_NEXT_SET(ip, whichfork,
5879                            XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5880         *logflags |= XFS_ILOG_CORE;
5881         if (!cur) {
5882                 *logflags |= XFS_ILOG_DEXT;
5883                 return 0;
5884         }
5885
5886         /* lookup and remove the extent to merge */
5887         error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5888                                    got.br_blockcount, &i);
5889         if (error)
5890                 return error;
5891         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5892
5893         error = xfs_btree_delete(cur, &i);
5894         if (error)
5895                 return error;
5896         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5897
5898         /* lookup and update size of the previous extent */
5899         error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock,
5900                                    left.br_blockcount, &i);
5901         if (error)
5902                 return error;
5903         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5904
5905         left.br_blockcount = blockcount;
5906
5907         return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock,
5908                                left.br_blockcount, left.br_state);
5909 }
5910
5911 /*
5912  * Shift a single extent.
5913  */
5914 STATIC int
5915 xfs_bmse_shift_one(
5916         struct xfs_inode                *ip,
5917         int                             whichfork,
5918         xfs_fileoff_t                   offset_shift_fsb,
5919         int                             *current_ext,
5920         struct xfs_bmbt_rec_host        *gotp,
5921         struct xfs_btree_cur            *cur,
5922         int                             *logflags,
5923         enum shift_direction            direction,
5924         struct xfs_defer_ops            *dfops)
5925 {
5926         struct xfs_ifork                *ifp;
5927         struct xfs_mount                *mp;
5928         xfs_fileoff_t                   startoff;
5929         struct xfs_bmbt_rec_host        *adj_irecp;
5930         struct xfs_bmbt_irec            got;
5931         struct xfs_bmbt_irec            adj_irec;
5932         int                             error;
5933         int                             i;
5934         int                             total_extents;
5935
5936         mp = ip->i_mount;
5937         ifp = XFS_IFORK_PTR(ip, whichfork);
5938         total_extents = xfs_iext_count(ifp);
5939
5940         xfs_bmbt_get_all(gotp, &got);
5941
5942         /* delalloc extents should be prevented by caller */
5943         XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock));
5944
5945         if (direction == SHIFT_LEFT) {
5946                 startoff = got.br_startoff - offset_shift_fsb;
5947
5948                 /*
5949                  * Check for merge if we've got an extent to the left,
5950                  * otherwise make sure there's enough room at the start
5951                  * of the file for the shift.
5952                  */
5953                 if (!*current_ext) {
5954                         if (got.br_startoff < offset_shift_fsb)
5955                                 return -EINVAL;
5956                         goto update_current_ext;
5957                 }
5958                 /*
5959                  * grab the left extent and check for a large
5960                  * enough hole.
5961                  */
5962                 adj_irecp = xfs_iext_get_ext(ifp, *current_ext - 1);
5963                 xfs_bmbt_get_all(adj_irecp, &adj_irec);
5964
5965                 if (startoff <
5966                     adj_irec.br_startoff + adj_irec.br_blockcount)
5967                         return -EINVAL;
5968
5969                 /* check whether to merge the extent or shift it down */
5970                 if (xfs_bmse_can_merge(&adj_irec, &got,
5971                                        offset_shift_fsb)) {
5972                         error = xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
5973                                                *current_ext, gotp, adj_irecp,
5974                                                cur, logflags);
5975                         if (error)
5976                                 return error;
5977                         adj_irec = got;
5978                         goto update_rmap;
5979                 }
5980         } else {
5981                 startoff = got.br_startoff + offset_shift_fsb;
5982                 /* nothing to move if this is the last extent */
5983                 if (*current_ext >= (total_extents - 1))
5984                         goto update_current_ext;
5985                 /*
5986                  * If this is not the last extent in the file, make sure there
5987                  * is enough room between current extent and next extent for
5988                  * accommodating the shift.
5989                  */
5990                 adj_irecp = xfs_iext_get_ext(ifp, *current_ext + 1);
5991                 xfs_bmbt_get_all(adj_irecp, &adj_irec);
5992                 if (startoff + got.br_blockcount > adj_irec.br_startoff)
5993                         return -EINVAL;
5994                 /*
5995                  * Unlike a left shift (which involves a hole punch),
5996                  * a right shift does not modify extent neighbors
5997                  * in any way. We should never find mergeable extents
5998                  * in this scenario. Check anyways and warn if we
5999                  * encounter two extents that could be one.
6000                  */
6001                 if (xfs_bmse_can_merge(&got, &adj_irec, offset_shift_fsb))
6002                         WARN_ON_ONCE(1);
6003         }
6004         /*
6005          * Increment the extent index for the next iteration, update the start
6006          * offset of the in-core extent and update the btree if applicable.
6007          */
6008 update_current_ext:
6009         if (direction == SHIFT_LEFT)
6010                 (*current_ext)++;
6011         else
6012                 (*current_ext)--;
6013         xfs_bmbt_set_startoff(gotp, startoff);
6014         *logflags |= XFS_ILOG_CORE;
6015         adj_irec = got;
6016         if (!cur) {
6017                 *logflags |= XFS_ILOG_DEXT;
6018                 goto update_rmap;
6019         }
6020
6021         error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
6022                                    got.br_blockcount, &i);
6023         if (error)
6024                 return error;
6025         XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
6026
6027         got.br_startoff = startoff;
6028         error = xfs_bmbt_update(cur, got.br_startoff, got.br_startblock,
6029                         got.br_blockcount, got.br_state);
6030         if (error)
6031                 return error;
6032
6033 update_rmap:
6034         /* update reverse mapping */
6035         error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, &adj_irec);
6036         if (error)
6037                 return error;
6038         adj_irec.br_startoff = startoff;
6039         return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &adj_irec);
6040 }
6041
6042 /*
6043  * Shift extent records to the left/right to cover/create a hole.
6044  *
6045  * The maximum number of extents to be shifted in a single operation is
6046  * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
6047  * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
6048  * is the length by which each extent is shifted. If there is no hole to shift
6049  * the extents into, this will be considered invalid operation and we abort
6050  * immediately.
6051  */
6052 int
6053 xfs_bmap_shift_extents(
6054         struct xfs_trans        *tp,
6055         struct xfs_inode        *ip,
6056         xfs_fileoff_t           *next_fsb,
6057         xfs_fileoff_t           offset_shift_fsb,
6058         int                     *done,
6059         xfs_fileoff_t           stop_fsb,
6060         xfs_fsblock_t           *firstblock,
6061         struct xfs_defer_ops    *dfops,
6062         enum shift_direction    direction,
6063         int                     num_exts)
6064 {
6065         struct xfs_btree_cur            *cur = NULL;
6066         struct xfs_bmbt_rec_host        *gotp;
6067         struct xfs_bmbt_irec            got;
6068         struct xfs_mount                *mp = ip->i_mount;
6069         struct xfs_ifork                *ifp;
6070         xfs_extnum_t                    nexts = 0;
6071         xfs_extnum_t                    current_ext;
6072         xfs_extnum_t                    total_extents;
6073         xfs_extnum_t                    stop_extent;
6074         int                             error = 0;
6075         int                             whichfork = XFS_DATA_FORK;
6076         int                             logflags = 0;
6077
6078         if (unlikely(XFS_TEST_ERROR(
6079             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
6080              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
6081              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
6082                 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
6083                                  XFS_ERRLEVEL_LOW, mp);
6084                 return -EFSCORRUPTED;
6085         }
6086
6087         if (XFS_FORCED_SHUTDOWN(mp))
6088                 return -EIO;
6089
6090         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
6091         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6092         ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
6093         ASSERT(*next_fsb != NULLFSBLOCK || direction == SHIFT_RIGHT);
6094
6095         ifp = XFS_IFORK_PTR(ip, whichfork);
6096         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6097                 /* Read in all the extents */
6098                 error = xfs_iread_extents(tp, ip, whichfork);
6099                 if (error)
6100                         return error;
6101         }
6102
6103         if (ifp->if_flags & XFS_IFBROOT) {
6104                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6105                 cur->bc_private.b.firstblock = *firstblock;
6106                 cur->bc_private.b.dfops = dfops;
6107                 cur->bc_private.b.flags = 0;
6108         }
6109
6110         /*
6111          * There may be delalloc extents in the data fork before the range we
6112          * are collapsing out, so we cannot use the count of real extents here.
6113          * Instead we have to calculate it from the incore fork.
6114          */
6115         total_extents = xfs_iext_count(ifp);
6116         if (total_extents == 0) {
6117                 *done = 1;
6118                 goto del_cursor;
6119         }
6120
6121         /*
6122          * In case of first right shift, we need to initialize next_fsb
6123          */
6124         if (*next_fsb == NULLFSBLOCK) {
6125                 gotp = xfs_iext_get_ext(ifp, total_extents - 1);
6126                 xfs_bmbt_get_all(gotp, &got);
6127                 *next_fsb = got.br_startoff;
6128                 if (stop_fsb > *next_fsb) {
6129                         *done = 1;
6130                         goto del_cursor;
6131                 }
6132         }
6133
6134         /* Lookup the extent index at which we have to stop */
6135         if (direction == SHIFT_RIGHT) {
6136                 gotp = xfs_iext_bno_to_ext(ifp, stop_fsb, &stop_extent);
6137                 /* Make stop_extent exclusive of shift range */
6138                 stop_extent--;
6139         } else
6140                 stop_extent = total_extents;
6141
6142         /*
6143          * Look up the extent index for the fsb where we start shifting. We can
6144          * henceforth iterate with current_ext as extent list changes are locked
6145          * out via ilock.
6146          *
6147          * gotp can be null in 2 cases: 1) if there are no extents or 2)
6148          * *next_fsb lies in a hole beyond which there are no extents. Either
6149          * way, we are done.
6150          */
6151         gotp = xfs_iext_bno_to_ext(ifp, *next_fsb, &current_ext);
6152         if (!gotp) {
6153                 *done = 1;
6154                 goto del_cursor;
6155         }
6156
6157         /* some sanity checking before we finally start shifting extents */
6158         if ((direction == SHIFT_LEFT && current_ext >= stop_extent) ||
6159              (direction == SHIFT_RIGHT && current_ext <= stop_extent)) {
6160                 error = -EIO;
6161                 goto del_cursor;
6162         }
6163
6164         while (nexts++ < num_exts) {
6165                 error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
6166                                            &current_ext, gotp, cur, &logflags,
6167                                            direction, dfops);
6168                 if (error)
6169                         goto del_cursor;
6170                 /*
6171                  * If there was an extent merge during the shift, the extent
6172                  * count can change. Update the total and grade the next record.
6173                  */
6174                 if (direction == SHIFT_LEFT) {
6175                         total_extents = xfs_iext_count(ifp);
6176                         stop_extent = total_extents;
6177                 }
6178
6179                 if (current_ext == stop_extent) {
6180                         *done = 1;
6181                         *next_fsb = NULLFSBLOCK;
6182                         break;
6183                 }
6184                 gotp = xfs_iext_get_ext(ifp, current_ext);
6185         }
6186
6187         if (!*done) {
6188                 xfs_bmbt_get_all(gotp, &got);
6189                 *next_fsb = got.br_startoff;
6190         }
6191
6192 del_cursor:
6193         if (cur)
6194                 xfs_btree_del_cursor(cur,
6195                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6196
6197         if (logflags)
6198                 xfs_trans_log_inode(tp, ip, logflags);
6199
6200         return error;
6201 }
6202
6203 /*
6204  * Splits an extent into two extents at split_fsb block such that it is
6205  * the first block of the current_ext. @current_ext is a target extent
6206  * to be split. @split_fsb is a block where the extents is split.
6207  * If split_fsb lies in a hole or the first block of extents, just return 0.
6208  */
6209 STATIC int
6210 xfs_bmap_split_extent_at(
6211         struct xfs_trans        *tp,
6212         struct xfs_inode        *ip,
6213         xfs_fileoff_t           split_fsb,
6214         xfs_fsblock_t           *firstfsb,
6215         struct xfs_defer_ops    *dfops)
6216 {
6217         int                             whichfork = XFS_DATA_FORK;
6218         struct xfs_btree_cur            *cur = NULL;
6219         struct xfs_bmbt_rec_host        *gotp;
6220         struct xfs_bmbt_irec            got;
6221         struct xfs_bmbt_irec            new; /* split extent */
6222         struct xfs_mount                *mp = ip->i_mount;
6223         struct xfs_ifork                *ifp;
6224         xfs_fsblock_t                   gotblkcnt; /* new block count for got */
6225         xfs_extnum_t                    current_ext;
6226         int                             error = 0;
6227         int                             logflags = 0;
6228         int                             i = 0;
6229
6230         if (unlikely(XFS_TEST_ERROR(
6231             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
6232              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
6233              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
6234                 XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
6235                                  XFS_ERRLEVEL_LOW, mp);
6236                 return -EFSCORRUPTED;
6237         }
6238
6239         if (XFS_FORCED_SHUTDOWN(mp))
6240                 return -EIO;
6241
6242         ifp = XFS_IFORK_PTR(ip, whichfork);
6243         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6244                 /* Read in all the extents */
6245                 error = xfs_iread_extents(tp, ip, whichfork);
6246                 if (error)
6247                         return error;
6248         }
6249
6250         /*
6251          * gotp can be null in 2 cases: 1) if there are no extents
6252          * or 2) split_fsb lies in a hole beyond which there are
6253          * no extents. Either way, we are done.
6254          */
6255         gotp = xfs_iext_bno_to_ext(ifp, split_fsb, &current_ext);
6256         if (!gotp)
6257                 return 0;
6258
6259         xfs_bmbt_get_all(gotp, &got);
6260
6261         /*
6262          * Check split_fsb lies in a hole or the start boundary offset
6263          * of the extent.
6264          */
6265         if (got.br_startoff >= split_fsb)
6266                 return 0;
6267
6268         gotblkcnt = split_fsb - got.br_startoff;
6269         new.br_startoff = split_fsb;
6270         new.br_startblock = got.br_startblock + gotblkcnt;
6271         new.br_blockcount = got.br_blockcount - gotblkcnt;
6272         new.br_state = got.br_state;
6273
6274         if (ifp->if_flags & XFS_IFBROOT) {
6275                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6276                 cur->bc_private.b.firstblock = *firstfsb;
6277                 cur->bc_private.b.dfops = dfops;
6278                 cur->bc_private.b.flags = 0;
6279                 error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
6280                                 got.br_startblock,
6281                                 got.br_blockcount,
6282                                 &i);
6283                 if (error)
6284                         goto del_cursor;
6285                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6286         }
6287
6288         xfs_bmbt_set_blockcount(gotp, gotblkcnt);
6289         got.br_blockcount = gotblkcnt;
6290
6291         logflags = XFS_ILOG_CORE;
6292         if (cur) {
6293                 error = xfs_bmbt_update(cur, got.br_startoff,
6294                                 got.br_startblock,
6295                                 got.br_blockcount,
6296                                 got.br_state);
6297                 if (error)
6298                         goto del_cursor;
6299         } else
6300                 logflags |= XFS_ILOG_DEXT;
6301
6302         /* Add new extent */
6303         current_ext++;
6304         xfs_iext_insert(ip, current_ext, 1, &new, 0);
6305         XFS_IFORK_NEXT_SET(ip, whichfork,
6306                            XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
6307
6308         if (cur) {
6309                 error = xfs_bmbt_lookup_eq(cur, new.br_startoff,
6310                                 new.br_startblock, new.br_blockcount,
6311                                 &i);
6312                 if (error)
6313                         goto del_cursor;
6314                 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
6315                 cur->bc_rec.b.br_state = new.br_state;
6316
6317                 error = xfs_btree_insert(cur, &i);
6318                 if (error)
6319                         goto del_cursor;
6320                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6321         }
6322
6323         /*
6324          * Convert to a btree if necessary.
6325          */
6326         if (xfs_bmap_needs_btree(ip, whichfork)) {
6327                 int tmp_logflags; /* partial log flag return val */
6328
6329                 ASSERT(cur == NULL);
6330                 error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, dfops,
6331                                 &cur, 0, &tmp_logflags, whichfork);
6332                 logflags |= tmp_logflags;
6333         }
6334
6335 del_cursor:
6336         if (cur) {
6337                 cur->bc_private.b.allocated = 0;
6338                 xfs_btree_del_cursor(cur,
6339                                 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6340         }
6341
6342         if (logflags)
6343                 xfs_trans_log_inode(tp, ip, logflags);
6344         return error;
6345 }
6346
6347 int
6348 xfs_bmap_split_extent(
6349         struct xfs_inode        *ip,
6350         xfs_fileoff_t           split_fsb)
6351 {
6352         struct xfs_mount        *mp = ip->i_mount;
6353         struct xfs_trans        *tp;
6354         struct xfs_defer_ops    dfops;
6355         xfs_fsblock_t           firstfsb;
6356         int                     error;
6357
6358         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
6359                         XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
6360         if (error)
6361                 return error;
6362
6363         xfs_ilock(ip, XFS_ILOCK_EXCL);
6364         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6365
6366         xfs_defer_init(&dfops, &firstfsb);
6367
6368         error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
6369                         &firstfsb, &dfops);
6370         if (error)
6371                 goto out;
6372
6373         error = xfs_defer_finish(&tp, &dfops, NULL);
6374         if (error)
6375                 goto out;
6376
6377         return xfs_trans_commit(tp);
6378
6379 out:
6380         xfs_defer_cancel(&dfops);
6381         xfs_trans_cancel(tp);
6382         return error;
6383 }
6384
6385 /* Deferred mapping is only for real extents in the data fork. */
6386 static bool
6387 xfs_bmap_is_update_needed(
6388         struct xfs_bmbt_irec    *bmap)
6389 {
6390         return  bmap->br_startblock != HOLESTARTBLOCK &&
6391                 bmap->br_startblock != DELAYSTARTBLOCK;
6392 }
6393
6394 /* Record a bmap intent. */
6395 static int
6396 __xfs_bmap_add(
6397         struct xfs_mount                *mp,
6398         struct xfs_defer_ops            *dfops,
6399         enum xfs_bmap_intent_type       type,
6400         struct xfs_inode                *ip,
6401         int                             whichfork,
6402         struct xfs_bmbt_irec            *bmap)
6403 {
6404         int                             error;
6405         struct xfs_bmap_intent          *bi;
6406
6407         trace_xfs_bmap_defer(mp,
6408                         XFS_FSB_TO_AGNO(mp, bmap->br_startblock),
6409                         type,
6410                         XFS_FSB_TO_AGBNO(mp, bmap->br_startblock),
6411                         ip->i_ino, whichfork,
6412                         bmap->br_startoff,
6413                         bmap->br_blockcount,
6414                         bmap->br_state);
6415
6416         bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS);
6417         INIT_LIST_HEAD(&bi->bi_list);
6418         bi->bi_type = type;
6419         bi->bi_owner = ip;
6420         bi->bi_whichfork = whichfork;
6421         bi->bi_bmap = *bmap;
6422
6423         error = xfs_defer_join(dfops, bi->bi_owner);
6424         if (error) {
6425                 kmem_free(bi);
6426                 return error;
6427         }
6428
6429         xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6430         return 0;
6431 }
6432
6433 /* Map an extent into a file. */
6434 int
6435 xfs_bmap_map_extent(
6436         struct xfs_mount        *mp,
6437         struct xfs_defer_ops    *dfops,
6438         struct xfs_inode        *ip,
6439         struct xfs_bmbt_irec    *PREV)
6440 {
6441         if (!xfs_bmap_is_update_needed(PREV))
6442                 return 0;
6443
6444         return __xfs_bmap_add(mp, dfops, XFS_BMAP_MAP, ip,
6445                         XFS_DATA_FORK, PREV);
6446 }
6447
6448 /* Unmap an extent out of a file. */
6449 int
6450 xfs_bmap_unmap_extent(
6451         struct xfs_mount        *mp,
6452         struct xfs_defer_ops    *dfops,
6453         struct xfs_inode        *ip,
6454         struct xfs_bmbt_irec    *PREV)
6455 {
6456         if (!xfs_bmap_is_update_needed(PREV))
6457                 return 0;
6458
6459         return __xfs_bmap_add(mp, dfops, XFS_BMAP_UNMAP, ip,
6460                         XFS_DATA_FORK, PREV);
6461 }
6462
6463 /*
6464  * Process one of the deferred bmap operations.  We pass back the
6465  * btree cursor to maintain our lock on the bmapbt between calls.
6466  */
6467 int
6468 xfs_bmap_finish_one(
6469         struct xfs_trans                *tp,
6470         struct xfs_defer_ops            *dfops,
6471         struct xfs_inode                *ip,
6472         enum xfs_bmap_intent_type       type,
6473         int                             whichfork,
6474         xfs_fileoff_t                   startoff,
6475         xfs_fsblock_t                   startblock,
6476         xfs_filblks_t                   blockcount,
6477         xfs_exntst_t                    state)
6478 {
6479         int                             error = 0, done;
6480
6481         trace_xfs_bmap_deferred(tp->t_mountp,
6482                         XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6483                         XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6484                         ip->i_ino, whichfork, startoff, blockcount, state);
6485
6486         if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK))
6487                 return -EFSCORRUPTED;
6488
6489         if (XFS_TEST_ERROR(false, tp->t_mountp,
6490                         XFS_ERRTAG_BMAP_FINISH_ONE,
6491                         XFS_RANDOM_BMAP_FINISH_ONE))
6492                 return -EIO;
6493
6494         switch (type) {
6495         case XFS_BMAP_MAP:
6496                 error = xfs_bmapi_remap(tp, ip, startoff, blockcount,
6497                                 startblock, dfops);
6498                 break;
6499         case XFS_BMAP_UNMAP:
6500                 error = xfs_bunmapi(tp, ip, startoff, blockcount,
6501                                 XFS_BMAPI_REMAP, 1, &startblock, dfops, &done);
6502                 ASSERT(done);
6503                 break;
6504         default:
6505                 ASSERT(0);
6506                 error = -EFSCORRUPTED;
6507         }
6508
6509         return error;
6510 }